release v1.6.4

.gitattributes

@@ -3,6 +3,7 @@ backend-python/wkv_cuda_utils/** linguist-vendored
 backend-python/get-pip.py linguist-vendored
 backend-python/convert_model.py linguist-vendored
 backend-python/convert_safetensors.py linguist-vendored
+backend-python/convert_pytorch_to_ggml.py linguist-vendored
 backend-python/utils/midi.py linguist-vendored
 build/** linguist-vendored
 finetune/lora/** linguist-vendored

.github/workflows/release.yml

@@ -48,16 +48,13 @@ jobs:
         id: cp310
         with:
           python-version: '3.10'
-      - uses: actions-rs/toolchain@v1
-        with:
-          toolchain: stable
-          override: true
-          target: wasm32-unknown-unknown
       - uses: crazy-max/ghaction-chocolatey@v2
         with:
           args: install upx
       - run: |
-          Start-BitsTransfer https://github.com/josStorer/LibreHardwareMonitor.Console/releases/download/v0.1.0/LibreHardwareMonitor.Console.zip ./LibreHardwareMonitor.Console.zip
+          Start-BitsTransfer https://github.com/josStorer/ai00_rwkv_server/releases/latest/download/webgpu_server_windows_x86_64.exe ./backend-rust/webgpu_server.exe
+          Start-BitsTransfer https://github.com/josStorer/web-rwkv-converter/releases/latest/download/web-rwkv-converter_windows_x86_64.exe ./backend-rust/web-rwkv-converter.exe
+          Start-BitsTransfer https://github.com/josStorer/LibreHardwareMonitor.Console/releases/latest/download/LibreHardwareMonitor.Console.zip ./LibreHardwareMonitor.Console.zip
           Expand-Archive ./LibreHardwareMonitor.Console.zip -DestinationPath ./components/LibreHardwareMonitor.Console
           Start-BitsTransfer https://www.python.org/ftp/python/3.10.11/python-3.10.11-embed-amd64.zip ./python-3.10.11-embed-amd64.zip
           Expand-Archive ./python-3.10.11-embed-amd64.zip -DestinationPath ./py310
@@ -67,13 +64,11 @@ jobs:
           Copy-Item -Path "${{ steps.cp310.outputs.python-path }}/../include" -Destination "py310/include" -Recurse
           Copy-Item -Path "${{ steps.cp310.outputs.python-path }}/../libs" -Destination "py310/libs" -Recurse
           ./py310/python -m pip install cyac==1.9
-          git clone https://github.com/josStorer/ai00_rwkv_server --depth=1
-          cd ai00_rwkv_server
-          cargo build --release
-          mv ./target/release/ai00_server.exe ../backend-rust/webgpu_server.exe
-          cd ..
           go install github.com/wailsapp/wails/v2/cmd/wails@latest
+          del ./backend-python/rwkv_pip/cpp/librwkv.dylib
+          del ./backend-python/rwkv_pip/cpp/librwkv.so
           (Get-Content -Path ./backend-golang/app.go) -replace "//go:custom_build windows ", "" | Set-Content -Path ./backend-golang/app.go
+          (Get-Content -Path ./backend-golang/utils.go) -replace "//go:custom_build windows ", "" | Set-Content -Path ./backend-golang/utils.go
           make
           Rename-Item -Path "build/bin/RWKV-Runner.exe" -NewName "RWKV-Runner_windows_x64.exe"
 
@@ -89,28 +84,21 @@ jobs:
       - uses: actions/setup-go@v4
         with:
           go-version: '1.20.5'
-      - uses: actions-rs/toolchain@v1
-        with:
-          toolchain: stable
-          override: true
-          target: wasm32-unknown-unknown
       - run: |
+          wget https://github.com/josStorer/ai00_rwkv_server/releases/latest/download/webgpu_server_linux_x86_64 -O ./backend-rust/webgpu_server
+          wget https://github.com/josStorer/web-rwkv-converter/releases/latest/download/web-rwkv-converter_linux_x86_64 -O ./backend-rust/web-rwkv-converter
           sudo apt-get update
           sudo apt-get install upx
-          sudo apt-get install build-essential libgtk-3-dev libwebkit2gtk-4.0-dev
-          git clone https://github.com/josStorer/ai00_rwkv_server --depth=1
-          cd ai00_rwkv_server
-          sudo apt-get install libudev-dev
-          sudo apt-get install libasound2-dev
-          rustup target add x86_64-unknown-linux-gnu
-          cargo build --release --target x86_64-unknown-linux-gnu
-          mv ./target/x86_64-unknown-linux-gnu/release/ai00_server ../backend-rust/webgpu_server
-          cd ..
+          sudo apt-get install build-essential libgtk-3-dev libwebkit2gtk-4.0-dev libasound2-dev
           go install github.com/wailsapp/wails/v2/cmd/wails@latest
           rm ./backend-python/rwkv_pip/wkv_cuda.pyd
           rm ./backend-python/rwkv_pip/rwkv5.pyd
+          rm ./backend-python/rwkv_pip/rwkv6.pyd
           rm ./backend-python/rwkv_pip/beta/wkv_cuda.pyd
           rm ./backend-python/get-pip.py
+          rm ./backend-python/rwkv_pip/cpp/librwkv.dylib
+          rm ./backend-python/rwkv_pip/cpp/rwkv.dll
+          rm ./backend-python/rwkv_pip/webgpu/web_rwkv_py.cp310-win_amd64.pyd
           make
           mv build/bin/RWKV-Runner build/bin/RWKV-Runner_linux_x64
 
@@ -126,22 +114,18 @@ jobs:
       - uses: actions/setup-go@v4
         with:
           go-version: '1.20.5'
-      - uses: actions-rs/toolchain@v1
-        with:
-          toolchain: stable
-          override: true
-          target: wasm32-unknown-unknown
       - run: |
-          git clone https://github.com/josStorer/ai00_rwkv_server --depth=1
-          cd ai00_rwkv_server
-          cargo build --release
-          mv ./target/release/ai00_server ../backend-rust/webgpu_server
-          cd ..
+          wget https://github.com/josStorer/ai00_rwkv_server/releases/latest/download/webgpu_server_darwin_aarch64 -O ./backend-rust/webgpu_server
+          wget https://github.com/josStorer/web-rwkv-converter/releases/latest/download/web-rwkv-converter_darwin_aarch64 -O ./backend-rust/web-rwkv-converter
           go install github.com/wailsapp/wails/v2/cmd/wails@latest
           rm ./backend-python/rwkv_pip/wkv_cuda.pyd
           rm ./backend-python/rwkv_pip/rwkv5.pyd
+          rm ./backend-python/rwkv_pip/rwkv6.pyd
           rm ./backend-python/rwkv_pip/beta/wkv_cuda.pyd
           rm ./backend-python/get-pip.py
+          rm ./backend-python/rwkv_pip/cpp/rwkv.dll
+          rm ./backend-python/rwkv_pip/cpp/librwkv.so
+          rm ./backend-python/rwkv_pip/webgpu/web_rwkv_py.cp310-win_amd64.pyd
           make
           cp build/darwin/Readme_Install.txt build/bin/Readme_Install.txt
           cp build/bin/RWKV-Runner.app/Contents/MacOS/RWKV-Runner build/bin/RWKV-Runner_darwin_universal

.gitignore

@@ -8,6 +8,7 @@ __pycache__
 *.st
 *.safetensors
 *.bin
+*.mid
 /config.json
 /cache.json
 /presets.json
@@ -18,7 +19,7 @@ __pycache__
 /cmd-helper.bat
 /install-py-dep.bat
 /backend-python/wkv_cuda
-/backend-python/rwkv5
+/backend-python/rwkv*
 *.exe
 *.old
 .DS_Store

CURRENT_CHANGE.md

@@ -1,14 +1,10 @@
 ## Changes
 
-### Improvements
-
-- improve mobile webview
-- add client upgrade progress
-- improve user guide
+- WebGPU (Python) strategy
 
 ## Install
 
 - Windows: https://github.com/josStorer/RWKV-Runner/blob/master/build/windows/Readme_Install.txt
 - MacOS: https://github.com/josStorer/RWKV-Runner/blob/master/build/darwin/Readme_Install.txt
 - Linux: https://github.com/josStorer/RWKV-Runner/blob/master/build/linux/Readme_Install.txt
-- Server-Deploy-Examples: https://github.com/josStorer/RWKV-Runner/tree/master/deploy-examples
+- Server-Deploy-Examples: https://github.com/josStorer/RWKV-Runner/tree/master/deploy-examples

Makefile

@@ -8,15 +8,15 @@ endif
 
 build-windows:
 	@echo ---- build for windows
-	wails build -upx -ldflags "-s -w" -platform windows/amd64
+	wails build -upx -ldflags '-s -w -extldflags "-static"' -platform windows/amd64
 
 build-macos:
 	@echo ---- build for macos
-	wails build -ldflags "-s -w" -platform darwin/universal
+	wails build -ldflags '-s -w' -platform darwin/universal
 
 build-linux:
 	@echo ---- build for linux
-	wails build -upx -ldflags "-s -w" -platform linux/amd64
+	wails build -upx -ldflags '-s -w' -platform linux/amd64
 
 build-web:
 	@echo ---- build for web

README.md

@@ -21,7 +21,7 @@ English | [简体中文](README_ZH.md) | [日本語](README_JA.md)
 [![MacOS][MacOS-image]][MacOS-url]
 [![Linux][Linux-image]][Linux-url]
 
-[FAQs](https://github.com/josStorer/RWKV-Runner/wiki/FAQs) | [Preview](#Preview) | [Download][download-url] | [Server-Deploy-Examples](https://github.com/josStorer/RWKV-Runner/tree/master/deploy-examples)
+[FAQs](https://github.com/josStorer/RWKV-Runner/wiki/FAQs) | [Preview](#Preview) | [Download][download-url] | [Simple Deploy Example](#Simple-Deploy-Example) | [Server Deploy Examples](https://github.com/josStorer/RWKV-Runner/tree/master/deploy-examples) | [MIDI Hardware Input](#MIDI-Input)
 
 [license-image]: http://img.shields.io/badge/license-MIT-blue.svg
 
@@ -57,20 +57,49 @@ English | [简体中文](README_ZH.md) | [日本語](README_JA.md)
 
 ## Features
 
-- RWKV model management and one-click startup
-- Fully compatible with the OpenAI API, making every ChatGPT client an RWKV client. After starting the model,
+- RWKV model management and one-click startup.
+- Front-end and back-end separation, if you don't want to use the client, also allows for separately deploying the
+  front-end service, or the back-end inference service, or the back-end inference service with a WebUI.
+  [Simple Deploy Example](#Simple-Deploy-Example) | [Server Deploy Examples](https://github.com/josStorer/RWKV-Runner/tree/master/deploy-examples)
+- Compatible with the OpenAI API, making every ChatGPT client an RWKV client. After starting the model,
   open http://127.0.0.1:8000/docs to view more details.
-- Automatic dependency installation, requiring only a lightweight executable program
-- Configs with 2G to 32G VRAM are included, works well on almost all computers
-- User-friendly chat and completion interaction interface included
-- Easy-to-understand and operate parameter configuration
-- Built-in model conversion tool
-- Built-in download management and remote model inspection
-- Built-in one-click LoRA Finetune
-- Can also be used as an OpenAI ChatGPT and GPT-Playground client
-- Multilingual localization
-- Theme switching
-- Automatic updates
+- Automatic dependency installation, requiring only a lightweight executable program.
+- Pre-set multi-level VRAM configs, works well on almost all computers. In Configs page, switch Strategy to WebGPU, it
+  can also run on AMD, Intel, and other graphics cards.
+- User-friendly chat, completion, and composition interaction interface included. Also supports chat presets, attachment
+  uploads, MIDI hardware input, and track editing.
+  [Preview](#Preview) | [MIDI Hardware Input](#MIDI-Input)
+- Built-in WebUI option, one-click start of Web service, sharing your hardware resources.
+- Easy-to-understand and operate parameter configuration, along with various operation guidance prompts.
+- Built-in model conversion tool.
+- Built-in download management and remote model inspection.
+- Built-in one-click LoRA Finetune. (Windows Only)
+- Can also be used as an OpenAI ChatGPT and GPT-Playground client. (Fill in the API URL and API Key in Settings page)
+- Multilingual localization.
+- Theme switching.
+- Automatic updates.
+
+## Simple Deploy Example
+
+```bash
+git clone https://github.com/josStorer/RWKV-Runner
+
+# Then
+cd RWKV-Runner
+python ./backend-python/main.py #The backend inference service has been started, request /switch-model API to load the model, refer to the API documentation: http://127.0.0.1:8000/docs
+
+# Or
+cd RWKV-Runner/frontend
+npm ci
+npm run build #Compile the frontend
+cd ..
+python ./backend-python/webui_server.py #Start the frontend service separately
+# Or
+python ./backend-python/main.py --webui #Start the frontend and backend service at the same time
+
+# Help Info
+python ./backend-python/main.py -h
+```
 
 ## API Concurrency Stress Testing
 
@@ -133,6 +162,48 @@ for i in np.argsort(embeddings_cos_sim)[::-1]:
     print(f"{embeddings_cos_sim[i]:.10f} - {values[i]}")
 ```
 
+## MIDI Input
+
+Tip: You can download https://github.com/josStorer/sgm_plus and unzip it to the program's `assets/sound-font` directory
+to use it as an offline sound source. Please note that if you are compiling the program from source code, do not place
+it in the source code directory.
+
+### USB MIDI Connection
+
+- USB MIDI devices are plug-and-play, and you can select your input device in the Composition page
+- ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/13bb92c3-4504-482d-ab82-026ac6c31095)
+
+### Mac MIDI Bluetooth Connection
+
+- For Mac users who want to use Bluetooth input,
+  please install [Bluetooth MIDI Connect](https://apps.apple.com/us/app/bluetooth-midi-connect/id1108321791), then click
+  the tray icon to connect after launching,
+  afterwards, you can select your input device in the Composition page.
+- ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/c079a109-1e3d-45c1-bbf5-eed85da1550e)
+
+### Windows MIDI Bluetooth Connection
+
+- Windows seems to have implemented Bluetooth MIDI support only for UWP (Universal Windows Platform) apps. Therefore, it
+  requires multiple steps to establish a connection. We need to create a local virtual MIDI device and then launch a UWP
+  application. Through this UWP application, we will redirect Bluetooth MIDI input to the virtual MIDI device, and then
+  this software will listen to the input from the virtual MIDI device.
+- So, first, you need to
+  download [loopMIDI](https://www.tobias-erichsen.de/wp-content/uploads/2020/01/loopMIDISetup_1_0_16_27.zip)
+  to create a virtual MIDI device. Click the plus sign in the bottom left corner to create the device.
+- ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/b75998ff-115c-4ddd-b97c-deeb5c106255)
+- Next, you need to download [Bluetooth LE Explorer](https://apps.microsoft.com/detail/9N0ZTKF1QD98) to discover and
+  connect to Bluetooth MIDI devices. Click "Start" to search for devices, and then click "Pair" to bind the MIDI device.
+- ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/c142c3ea-a973-4531-9807-4c385d640a2b)
+- Finally, you need to install [MIDIberry](https://apps.microsoft.com/detail/9N39720H2M05),
+  This UWP application can redirect Bluetooth MIDI input to the virtual MIDI device. After launching it, double-click
+  your actual Bluetooth MIDI device name in the input field, and in the output field, double-click the virtual MIDI
+  device name we created earlier.
+- ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/5ad6a1d9-4f68-4d95-ae17-4296107d1669)
+- Now, you can select the virtual MIDI device as the input in the Composition page. Bluetooth LE Explorer no longer
+  needs to run, and you can also close the loopMIDI window, it will run automatically in the background. Just keep
+  MIDIberry open.
+- ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/1c371821-c7b7-4c18-8e42-9e315efbe427)
+
 ## Related Repositories:
 
 - RWKV-4-World: https://huggingface.co/BlinkDL/rwkv-4-world/tree/main
@@ -146,27 +217,35 @@ for i in np.argsort(embeddings_cos_sim)[::-1]:
 
 ### Homepage
 
-![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/d7f24d80-f382-428d-8b28-edf87e1549e2)
+![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/c9b9cdd0-63f9-4319-9f74-5bf5d7df5a67)
 
 ### Chat
 
 ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/80009872-528f-4932-aeb2-f724fa892e7c)
 
+![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/e98c9038-3323-47b0-8edb-d639fafd37b2)
+
 ### Completion
 
 ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/bf49de8e-3b89-4543-b1ef-7cd4b19a1836)
 
 ### Composition
 
+Tip: You can download https://github.com/josStorer/sgm_plus and unzip it to the program's `assets/sound-font` directory
+to use it as an offline sound source. Please note that if you are compiling the program from source code, do not place
+it in the source code directory.
+
 ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/e8ad908d-3fd2-4e92-bcdb-96815cb836ee)
 
+![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/b2ce4761-9e75-477e-a182-d0255fb8ac76)
+
 ### Configuration
 
-![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/48befdc6-e03c-4851-9bee-22f77ee2640e)
+![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/f41060dc-5517-44af-bb3f-8ef71720016d)
 
 ### Model Management
 
-![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/367fe4f8-cc12-475f-9371-3cf62cdbf293)
+![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/b1581147-a6ce-4493-8010-e33c0ddeca0a)
 
 ### Download Management
 

README_JA.md

@@ -21,7 +21,7 @@
 [![MacOS][MacOS-image]][MacOS-url]
 [![Linux][Linux-image]][Linux-url]
 
-[FAQs](https://github.com/josStorer/RWKV-Runner/wiki/FAQs) | [プレビュー](#Preview) | [ダウンロード][download-url] | [サーバーデプロイ例](https://github.com/josStorer/RWKV-Runner/tree/master/deploy-examples)
+[FAQs](https://github.com/josStorer/RWKV-Runner/wiki/FAQs) | [プレビュー](#Preview) | [ダウンロード][download-url] | [シンプルなデプロイの例](#Simple-Deploy-Example) | [サーバーデプロイ例](https://github.com/josStorer/RWKV-Runner/tree/master/deploy-examples) | [MIDIハードウェア入力](#MIDI-Input)
 
 [license-image]: http://img.shields.io/badge/license-MIT-blue.svg
 
@@ -58,20 +58,47 @@
 ## 特徴
 
 - RWKV モデル管理とワンクリック起動
-- OpenAI API と完全に互換性があり、すべての ChatGPT クライアントを RWKV クライアントにします。モデル起動後、
+- フロントエンドとバックエンドの分離は、クライアントを使用しない場合でも、フロントエンドサービス、またはバックエンド推論サービス、またはWebUIを備えたバックエンド推論サービスを個別に展開することを可能にします。
+  [シンプルなデプロイの例](#Simple-Deploy-Example) | [サーバーデプロイ例](https://github.com/josStorer/RWKV-Runner/tree/master/deploy-examples)
+- OpenAI API と互換性があり、すべての ChatGPT クライアントを RWKV クライアントにします。モデル起動後、
   http://127.0.0.1:8000/docs を開いて詳細をご覧ください。
 - 依存関係の自動インストールにより、軽量な実行プログラムのみを必要とします
-- 2G から 32G の VRAM のコンフィグが含まれており、ほとんどのコンピュータで動作します
-- ユーザーフレンドリーなチャットと完成インタラクションインターフェースを搭載
-- 分かりやすく操作しやすいパラメータ設定
+- 事前設定された多段階のVRAM設定、ほとんどのコンピュータで動作します。配置ページで、ストラテジーをWebGPUに切り替えると、AMD、インテル、その他のグラフィックカードでも動作します
+- ユーザーフレンドリーなチャット、完成、および作曲インターフェイスが含まれています。また、チャットプリセット、添付ファイルのアップロード、MIDIハードウェア入力、トラック編集もサポートしています。
+  [プレビュー](#Preview) | [MIDIハードウェア入力](#MIDI-Input)
+- 内蔵WebUIオプション、Webサービスのワンクリック開始、ハードウェアリソースの共有
+- 分かりやすく操作しやすいパラメータ設定、各種操作ガイダンスプロンプトとともに
 - 内蔵モデル変換ツール
 - ダウンロード管理とリモートモデル検査機能内蔵
-- 内蔵のLoRA微調整機能を搭載しています
-- このプログラムは、OpenAI ChatGPTとGPT Playgroundのクライアントとしても使用できます
+- 内蔵のLoRA微調整機能を搭載しています (Windowsのみ)
+- このプログラムは、OpenAI ChatGPTとGPT Playgroundのクライアントとしても使用できます（設定ページで `API URL` と `API Key`
+  を入力してください）
 - 多言語ローカライズ
 - テーマ切り替え
 - 自動アップデート
 
+## Simple Deploy Example
+
+```bash
+git clone https://github.com/josStorer/RWKV-Runner
+
+# Then
+cd RWKV-Runner
+python ./backend-python/main.py #The backend inference service has been started, request /switch-model API to load the model, refer to the API documentation: http://127.0.0.1:8000/docs
+
+# Or
+cd RWKV-Runner/frontend
+npm ci
+npm run build #Compile the frontend
+cd ..
+python ./backend-python/webui_server.py #Start the frontend service separately
+# Or
+python ./backend-python/main.py --webui #Start the frontend and backend service at the same time
+
+# Help Info
+python ./backend-python/main.py -h
+```
+
 ## API 同時実行ストレステスト
 
 ```bash
@@ -134,6 +161,48 @@ for i in np.argsort(embeddings_cos_sim)[::-1]:
     print(f"{embeddings_cos_sim[i]:.10f} - {values[i]}")
 ```
 
+## MIDI Input
+
+Tip: You can download https://github.com/josStorer/sgm_plus and unzip it to the program's `assets/sound-font` directory
+to use it as an offline sound source. Please note that if you are compiling the program from source code, do not place
+it in the source code directory.
+
+### USB MIDI Connection
+
+- USB MIDI devices are plug-and-play, and you can select your input device in the Composition page
+- ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/13bb92c3-4504-482d-ab82-026ac6c31095)
+
+### Mac MIDI Bluetooth Connection
+
+- For Mac users who want to use Bluetooth input,
+  please install [Bluetooth MIDI Connect](https://apps.apple.com/us/app/bluetooth-midi-connect/id1108321791), then click
+  the tray icon to connect after launching,
+  afterwards, you can select your input device in the Composition page.
+- ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/c079a109-1e3d-45c1-bbf5-eed85da1550e)
+
+### Windows MIDI Bluetooth Connection
+
+- Windows seems to have implemented Bluetooth MIDI support only for UWP (Universal Windows Platform) apps. Therefore, it
+  requires multiple steps to establish a connection. We need to create a local virtual MIDI device and then launch a UWP
+  application. Through this UWP application, we will redirect Bluetooth MIDI input to the virtual MIDI device, and then
+  this software will listen to the input from the virtual MIDI device.
+- So, first, you need to
+  download [loopMIDI](https://www.tobias-erichsen.de/wp-content/uploads/2020/01/loopMIDISetup_1_0_16_27.zip)
+  to create a virtual MIDI device. Click the plus sign in the bottom left corner to create the device.
+- ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/b75998ff-115c-4ddd-b97c-deeb5c106255)
+- Next, you need to download [Bluetooth LE Explorer](https://apps.microsoft.com/detail/9N0ZTKF1QD98) to discover and
+  connect to Bluetooth MIDI devices. Click "Start" to search for devices, and then click "Pair" to bind the MIDI device.
+- ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/c142c3ea-a973-4531-9807-4c385d640a2b)
+- Finally, you need to install [MIDIberry](https://apps.microsoft.com/detail/9N39720H2M05),
+  This UWP application can redirect Bluetooth MIDI input to the virtual MIDI device. After launching it, double-click
+  your actual Bluetooth MIDI device name in the input field, and in the output field, double-click the virtual MIDI
+  device name we created earlier.
+- ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/5ad6a1d9-4f68-4d95-ae17-4296107d1669)
+- Now, you can select the virtual MIDI device as the input in the Composition page. Bluetooth LE Explorer no longer
+  needs to run, and you can also close the loopMIDI window, it will run automatically in the background. Just keep
+  MIDIberry open.
+- ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/1c371821-c7b7-4c18-8e42-9e315efbe427)
+
 ## 関連リポジトリ:
 
 - RWKV-4-World: https://huggingface.co/BlinkDL/rwkv-4-world/tree/main
@@ -143,31 +212,39 @@ for i in np.argsort(embeddings_cos_sim)[::-1]:
 - RWKV-LM-LoRA: https://github.com/Blealtan/RWKV-LM-LoRA
 - MIDI-LLM-tokenizer: https://github.com/briansemrau/MIDI-LLM-tokenizer
 
-## プレビュー
+## Preview
 
 ### ホームページ
 
-![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/d7f24d80-f382-428d-8b28-edf87e1549e2)
+![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/c9b9cdd0-63f9-4319-9f74-5bf5d7df5a67)
 
 ### チャット
 
 ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/80009872-528f-4932-aeb2-f724fa892e7c)
 
+![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/e98c9038-3323-47b0-8edb-d639fafd37b2)
+
 ### 補完
 
 ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/bf49de8e-3b89-4543-b1ef-7cd4b19a1836)
 
 ### 作曲
 
+Tip: You can download https://github.com/josStorer/sgm_plus and unzip it to the program's `assets/sound-font` directory
+to use it as an offline sound source. Please note that if you are compiling the program from source code, do not place
+it in the source code directory.
+
 ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/e8ad908d-3fd2-4e92-bcdb-96815cb836ee)
 
+![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/b2ce4761-9e75-477e-a182-d0255fb8ac76)
+
 ### コンフィグ
 
-![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/48befdc6-e03c-4851-9bee-22f77ee2640e)
+![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/f41060dc-5517-44af-bb3f-8ef71720016d)
 
 ### モデル管理
 
-![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/367fe4f8-cc12-475f-9371-3cf62cdbf293)
+![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/b1581147-a6ce-4493-8010-e33c0ddeca0a)
 
 ### ダウンロード管理
 

README_ZH.md

@@ -20,7 +20,7 @@ API兼容的接口，这意味着一切ChatGPT客户端都是RWKV客户端。
 [![MacOS][MacOS-image]][MacOS-url]
 [![Linux][Linux-image]][Linux-url]
 
-[视频演示](https://www.bilibili.com/video/BV1hM4y1v76R) | [疑难解答](https://www.bilibili.com/read/cv23921171) | [预览](#Preview) | [下载][download-url] | [懒人包](https://pan.baidu.com/s/1zdzZ_a0uM3gDqi6pXIZVAA?pwd=1111) | [服务器部署示例](https://github.com/josStorer/RWKV-Runner/tree/master/deploy-examples)
+[视频演示](https://www.bilibili.com/video/BV1hM4y1v76R) | [疑难解答](https://www.bilibili.com/read/cv23921171) | [预览](#Preview) | [下载][download-url] | [懒人包](https://pan.baidu.com/s/1zdzZ_a0uM3gDqi6pXIZVAA?pwd=1111) | [简明服务部署示例](#Simple-Deploy-Example) | [服务器部署示例](https://github.com/josStorer/RWKV-Runner/tree/master/deploy-examples) | [MIDI硬件输入](#MIDI-Input)
 
 [license-image]: http://img.shields.io/badge/license-MIT-blue.svg
 
@@ -57,19 +57,45 @@ API兼容的接口，这意味着一切ChatGPT客户端都是RWKV客户端。
 ## 功能
 
 - RWKV模型管理，一键启动
-- 与OpenAI API完全兼容，一切ChatGPT客户端，都是RWKV客户端。启动模型后，打开 http://127.0.0.1:8000/docs 查看详细内容
+- 前后端分离，如果你不想使用客户端，也允许单独部署前端服务，或后端推理服务，或具有WebUI的后端推理服务。
+  [简明服务部署示例](#Simple-Deploy-Example) | [服务器部署示例](https://github.com/josStorer/RWKV-Runner/tree/master/deploy-examples)
+- 与OpenAI API兼容，一切ChatGPT客户端，都是RWKV客户端。启动模型后，打开 http://127.0.0.1:8000/docs 查看API文档
 - 全自动依赖安装，你只需要一个轻巧的可执行程序
-- 预设了2G至32G显存的配置，几乎在各种电脑上工作良好
-- 自带用户友好的聊天和续写交互页面
-- 易于理解和操作的参数配置
+- 预设多级显存配置，几乎在各种电脑上工作良好。通过配置页面切换Strategy到WebGPU，还可以在AMD，Intel等显卡上运行
+- 自带用户友好的聊天，续写，作曲交互页面。支持聊天预设，附件上传，MIDI硬件输入及音轨编辑。
+  [预览](#Preview) | [MIDI硬件输入](#MIDI-Input)
+- 内置WebUI选项，一键启动Web服务，共享硬件资源
+- 易于理解和操作的参数配置，及各类操作引导提示
 - 内置模型转换工具
 - 内置下载管理和远程模型检视
-- 内置一键LoRA微调
-- 也可用作 OpenAI ChatGPT 和 GPT Playground 客户端
+- 内置一键LoRA微调 (仅限Windows)
+- 也可用作 OpenAI ChatGPT 和 GPT Playground 客户端 (在设置内填写API URL和API Key)
 - 多语言本地化
 - 主题切换
 - 自动更新
 
+## Simple Deploy Example
+
+```bash
+git clone https://github.com/josStorer/RWKV-Runner
+
+# 然后
+cd RWKV-Runner
+python ./backend-python/main.py #后端推理服务已启动, 调用/switch-model载入模型, 参考API文档: http://127.0.0.1:8000/docs
+
+# 或者
+cd RWKV-Runner/frontend
+npm ci
+npm run build #编译前端
+cd ..
+python ./backend-python/webui_server.py #单独启动前端服务
+# 或者
+python ./backend-python/main.py --webui #同时启动前后端服务
+
+# 帮助参数
+python ./backend-python/main.py -h
+```
+
 ## API并发压力测试
 
 ```bash
@@ -130,6 +156,40 @@ for i in np.argsort(embeddings_cos_sim)[::-1]:
     print(f"{embeddings_cos_sim[i]:.10f} - {values[i]}")
 ```
 
+## MIDI Input
+
+小贴士: 你可以下载 https://github.com/josStorer/sgm_plus, 并解压到程序的`assets/sound-font`目录, 以使用离线音源. 注意,
+如果你正在从源码编译程序, 请不要将其放置在源码目录中
+
+### USB MIDI 连接
+
+- USB MIDI设备是即插即用的, 你能够在作曲页面选择你的输入设备
+- ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/a448c34a-56d8-46eb-8dc2-dd11e8e0c4ce)
+
+### Mac MIDI 蓝牙连接
+
+- 对于想要使用蓝牙输入的Mac用户,
+  请安装[Bluetooth MIDI Connect](https://apps.apple.com/us/app/bluetooth-midi-connect/id1108321791), 启动后点击托盘连接,
+  之后你可以在作曲页面选择你的输入设备
+- ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/c079a109-1e3d-45c1-bbf5-eed85da1550e)
+
+### Windows MIDI 蓝牙连接
+
+- Windows似乎只为UWP实现了蓝牙MIDI支持, 因此需要多个步骤进行连接, 我们需要创建一个本地的虚拟MIDI设备, 然后启动一个UWP应用,
+  通过此UWP应用将蓝牙MIDI输入重定向到虚拟MIDI设备, 然后本软件监听虚拟MIDI设备的输入
+- 因此, 首先你需要下载[loopMIDI](https://www.tobias-erichsen.de/wp-content/uploads/2020/01/loopMIDISetup_1_0_16_27.zip),
+  用于创建虚拟MIDI设备, 点击左下角的加号创建设备
+- ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/b75998ff-115c-4ddd-b97c-deeb5c106255)
+- 然后, 你需要下载[Bluetooth LE Explorer](https://apps.microsoft.com/detail/9N0ZTKF1QD98), 以发现并连接蓝牙MIDI设备,
+  点击Start搜索设备, 然后点击Pair绑定MIDI设备
+- ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/c142c3ea-a973-4531-9807-4c385d640a2b)
+- 最后, 你需要安装[MIDIberry](https://apps.microsoft.com/detail/9N39720H2M05), 这个UWP应用能将MIDI蓝牙输入重定向到虚拟MIDI设备,
+  启动后, 在输入栏, 双击你实际的蓝牙MIDI设备名称, 在输出栏, 双击我们先前创建的虚拟MIDI设备名称
+- ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/5ad6a1d9-4f68-4d95-ae17-4296107d1669)
+- 现在, 你可以在作曲页面选择虚拟MIDI设备作为输入. Bluetooth LE Explorer不再需要运行, loopMIDI窗口也可以退出, 它会自动在后台运行,
+  仅保持MIDIberry打开即可
+- ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/6460c355-884e-4b28-a2eb-8ab7a2e3a01a)
+
 ## 相关仓库:
 
 - RWKV-4-World: https://huggingface.co/BlinkDL/rwkv-4-world/tree/main
@@ -143,27 +203,34 @@ for i in np.argsort(embeddings_cos_sim)[::-1]:
 
 ### 主页
 
-![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/ff2b1eef-dd3b-4cbf-98fb-b5a1ecee43e1)
+![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/cd82674e-3ee3-4175-bd9c-a11d45437327)
 
 ### 聊天
 
 ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/9570e73b-dca2-4316-9e92-09961f3c48c4)
 
+![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/54bb0e2b-cdc4-4ea0-8d16-9beaf57c232c)
+
 ### 续写
 
 ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/69f9ba7a-2fe8-4a5e-94cb-aa655aa409e2)
 
 ### 作曲
 
+小贴士: 你可以下载 https://github.com/josStorer/sgm_plus, 并解压到程序的`assets/sound-font`目录, 以使用离线音源. 注意,
+如果你正在从源码编译程序, 请不要将其放置在源码目录中
+
 ![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/95b34893-80c2-4706-87f9-bc141032ed4b)
 
+![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/3cb31ca8-d708-42f1-8768-1605fb0b2174)
+
 ### 配置
 
-![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/59460f69-b172-4c7a-86cb-573262543076)
+![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/0f4d4f21-8abe-4f4d-8c4f-cd7d5607f20e)
 
 ### 模型管理
 
-![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/551121ee-1bfe-421b-a9d1-24125126ab4b)
+![image](https://github.com/josStorer/RWKV-Runner/assets/13366013/871f2d2a-7e41-4be7-9b32-be1b3e00dc3e)
 
 ### 下载管理
 

backend-golang/app.go

@@ -45,7 +45,8 @@ func (a *App) OnStartup(ctx context.Context) {
 		a.cmdPrefix = "cd " + a.exDir + " && "
 	}
 
-	os.Chmod("./backend-rust/webgpu_server", 0777)
+	os.Chmod(a.exDir+"backend-rust/webgpu_server", 0777)
+	os.Chmod(a.exDir+"backend-rust/web-rwkv-converter", 0777)
 	os.Mkdir(a.exDir+"models", os.ModePerm)
 	os.Mkdir(a.exDir+"lora-models", os.ModePerm)
 	os.Mkdir(a.exDir+"finetune/json2binidx_tool/data", os.ModePerm)
@@ -55,6 +56,7 @@ func (a *App) OnStartup(ctx context.Context) {
 	}
 
 	a.downloadLoop()
+	a.midiLoop()
 	a.watchFs()
 	a.monitorHardware()
 }
@@ -69,8 +71,8 @@ func (a *App) OnBeforeClose(ctx context.Context) bool {
 func (a *App) watchFs() {
 	watcher, err := fsnotify.NewWatcher()
 	if err == nil {
-		watcher.Add("./lora-models")
-		watcher.Add("./models")
+		watcher.Add(a.exDir + "./lora-models")
+		watcher.Add(a.exDir + "./models")
 		go func() {
 			for {
 				select {

backend-golang/download.go

@@ -33,9 +33,9 @@ type DownloadStatus struct {
 
 var downloadList []*DownloadStatus
 
-func existsInDownloadList(url string) bool {
+func existsInDownloadList(path string, url string) bool {
 	for _, ds := range downloadList {
-		if ds.Url == url {
+		if ds.Path == path || ds.Url == url {
 			return true
 		}
 	}
@@ -88,7 +88,7 @@ func (a *App) ContinueDownload(url string) {
 }
 
 func (a *App) AddToDownloadList(path string, url string) {
-	if !existsInDownloadList(url) {
+	if !existsInDownloadList(a.exDir+path, url) {
 		downloadList = append(downloadList, &DownloadStatus{
 			resp:        nil,
 			Name:        filepath.Base(path),

backend-golang/file.go

@@ -14,6 +14,13 @@ import (
 	wruntime "github.com/wailsapp/wails/v2/pkg/runtime"
 )
 
+func (a *App) SaveFile(path string, savedContent []byte) error {
+	if err := os.WriteFile(a.exDir+path, savedContent, 0644); err != nil {
+		return err
+	}
+	return nil
+}
+
 func (a *App) SaveJson(fileName string, jsonData any) error {
 	text, err := json.MarshalIndent(jsonData, "", "  ")
 	if err != nil {
@@ -195,3 +202,12 @@ func (a *App) OpenFileFolder(path string, relative bool) error {
 	}
 	return errors.New("unsupported OS")
 }
+
+func (a *App) StartFile(path string) error {
+	cmd, err := CmdHelper(true, path)
+	if err != nil {
+		return err
+	}
+	err = cmd.Start()
+	return err
+}

backend-golang/midi.go

@@ -0,0 +1,170 @@
+package backend_golang
+
+import (
+	"errors"
+	"fmt"
+	"time"
+
+	"github.com/mattrtaylor/go-rtmidi"
+	"github.com/wailsapp/wails/v2/pkg/runtime"
+)
+
+type Port struct {
+	Name string `json:"name"`
+}
+type MIDIMessage struct {
+	MessageType string `json:"messageType"`
+	Channel     int    `json:"channel"`
+	Note        int    `json:"note"`
+	Velocity    int    `json:"velocity"`
+	Control     int    `json:"control"`
+	Value       int    `json:"value"`
+}
+
+var ports []Port
+var input rtmidi.MIDIIn
+var out rtmidi.MIDIOut
+var activeIndex int = -1
+var lastNoteTime time.Time
+
+func (a *App) midiLoop() {
+	var err error
+	input, err = rtmidi.NewMIDIInDefault()
+	if err != nil {
+		runtime.EventsEmit(a.ctx, "midiError", err.Error())
+		return
+	}
+	out, err = rtmidi.NewMIDIOutDefault()
+	if err != nil {
+		runtime.EventsEmit(a.ctx, "midiError", err.Error())
+	}
+	err = out.OpenPort(0, "")
+	if err != nil {
+		runtime.EventsEmit(a.ctx, "midiError", err.Error())
+	}
+	ticker := time.NewTicker(500 * time.Millisecond)
+	go func() {
+		for {
+			<-ticker.C
+			count, err := input.PortCount()
+			if err != nil {
+				continue
+			}
+			ports = make([]Port, count)
+			for i := 0; i < count; i++ {
+				name, err := input.PortName(i)
+				if err == nil {
+					ports[i].Name = name
+				}
+			}
+			runtime.EventsEmit(a.ctx, "midiPorts", &ports)
+		}
+	}()
+}
+
+func (a *App) OpenMidiPort(index int) error {
+	if input == nil {
+		return errors.New("failed to initialize MIDI input")
+	}
+	if activeIndex == index {
+		return nil
+	}
+	input.Destroy()
+	var err error
+	input, err = rtmidi.NewMIDIInDefault()
+	if err != nil {
+		return err
+	}
+	err = input.SetCallback(func(msg rtmidi.MIDIIn, bytes []byte, t float64) {
+		// https://www.midi.org/specifications-old/item/table-1-summary-of-midi-message
+		// https://www.rfc-editor.org/rfc/rfc6295.html
+		//
+		// msgType channel
+		//  1001     0000
+		//
+		msgType := bytes[0] >> 4
+		channel := bytes[0] & 0x0f
+		switch msgType {
+		case 0x8:
+			elapsed := time.Since(lastNoteTime)
+			lastNoteTime = time.Now()
+			runtime.EventsEmit(a.ctx, "midiMessage", &MIDIMessage{
+				MessageType: "ElapsedTime",
+				Value:       int(elapsed.Milliseconds()),
+			})
+			note := bytes[1]
+			runtime.EventsEmit(a.ctx, "midiMessage", &MIDIMessage{
+				MessageType: "NoteOff",
+				Channel:     int(channel),
+				Note:        int(note),
+			})
+		case 0x9:
+			elapsed := time.Since(lastNoteTime)
+			lastNoteTime = time.Now()
+			runtime.EventsEmit(a.ctx, "midiMessage", &MIDIMessage{
+				MessageType: "ElapsedTime",
+				Value:       int(elapsed.Milliseconds()),
+			})
+			note := bytes[1]
+			velocity := bytes[2]
+			runtime.EventsEmit(a.ctx, "midiMessage", &MIDIMessage{
+				MessageType: "NoteOn",
+				Channel:     int(channel),
+				Note:        int(note),
+				Velocity:    int(velocity),
+			})
+		case 0xb:
+			// control 12 => K1 knob, control 13 => K2 knob
+			control := bytes[1]
+			value := bytes[2]
+			runtime.EventsEmit(a.ctx, "midiMessage", &MIDIMessage{
+				MessageType: "ControlChange",
+				Channel:     int(channel),
+				Control:     int(control),
+				Value:       int(value),
+			})
+		default:
+			fmt.Printf("Unknown midi message: %v\n", bytes)
+		}
+	})
+	if err != nil {
+		return err
+	}
+	err = input.OpenPort(index, "")
+	if err != nil {
+		return err
+	}
+	activeIndex = index
+	lastNoteTime = time.Now()
+	return nil
+}
+
+func (a *App) CloseMidiPort() error {
+	if input == nil {
+		return errors.New("failed to initialize MIDI input")
+	}
+	if activeIndex == -1 {
+		return nil
+	}
+	activeIndex = -1
+	input.Destroy()
+	var err error
+	input, err = rtmidi.NewMIDIInDefault()
+	if err != nil {
+		return err
+	}
+	return nil
+}
+
+func (a *App) PlayNote(msg MIDIMessage) error {
+	if out == nil {
+		return errors.New("failed to initialize MIDI output")
+	}
+	channelByte := byte(msg.Channel)
+	if msg.MessageType == "NoteOn" {
+		out.SendMessage([]byte{0x90 | channelByte, byte(msg.Note), byte(msg.Velocity)})
+	} else if msg.MessageType == "NoteOff" {
+		out.SendMessage([]byte{0x80 | channelByte, byte(msg.Note), byte(msg.Velocity)})
+	}
+	return nil
+}

backend-golang/rwkv.go

@@ -10,7 +10,7 @@ import (
 	"strings"
 )
 
-func (a *App) StartServer(python string, port int, host string, webui bool, rwkvBeta bool) (string, error) {
+func (a *App) StartServer(python string, port int, host string, webui bool, rwkvBeta bool, rwkvcpp bool, webgpu bool) (string, error) {
 	var err error
 	if python == "" {
 		python, err = GetPython()
@@ -25,6 +25,12 @@ func (a *App) StartServer(python string, port int, host string, webui bool, rwkv
 	if rwkvBeta {
 		args = append(args, "--rwkv-beta")
 	}
+	if rwkvcpp {
+		args = append(args, "--rwkv.cpp")
+	}
+	if webgpu {
+		args = append(args, "--webgpu")
+	}
 	args = append(args, "--port", strconv.Itoa(port), "--host", host)
 	return Cmd(args...)
 }
@@ -46,7 +52,13 @@ func (a *App) ConvertModel(python string, modelPath string, strategy string, out
 	return Cmd(python, "./backend-python/convert_model.py", "--in", modelPath, "--out", outPath, "--strategy", strategy)
 }
 
-func (a *App) ConvertSafetensors(python string, modelPath string, outPath string) (string, error) {
+func (a *App) ConvertSafetensors(modelPath string, outPath string) (string, error) {
+	args := []string{"./backend-rust/web-rwkv-converter"}
+	args = append(args, "--input", modelPath, "--output", outPath)
+	return Cmd(args...)
+}
+
+func (a *App) ConvertSafetensorsWithPython(python string, modelPath string, outPath string) (string, error) {
 	var err error
 	if python == "" {
 		python, err = GetPython()
@@ -57,6 +69,21 @@ func (a *App) ConvertSafetensors(python string, modelPath string, outPath string
 	return Cmd(python, "./backend-python/convert_safetensors.py", "--input", modelPath, "--output", outPath)
 }
 
+func (a *App) ConvertGGML(python string, modelPath string, outPath string, Q51 bool) (string, error) {
+	var err error
+	if python == "" {
+		python, err = GetPython()
+	}
+	if err != nil {
+		return "", err
+	}
+	dataType := "FP16"
+	if Q51 {
+		dataType = "Q5_1"
+	}
+	return Cmd(python, "./backend-python/convert_pytorch_to_ggml.py", modelPath, outPath, dataType)
+}
+
 func (a *App) ConvertData(python string, input string, outputPrefix string, vocab string) (string, error) {
 	var err error
 	if python == "" {

backend-golang/utils.go

@@ -15,33 +15,50 @@ import (
 	"runtime"
 	"strconv"
 	"strings"
+	"syscall"
 )
 
+func CmdHelper(hideWindow bool, args ...string) (*exec.Cmd, error) {
+	if runtime.GOOS != "windows" {
+		return nil, errors.New("unsupported OS")
+	}
+	filename := "./cmd-helper.bat"
+	_, err := os.Stat(filename)
+	if err != nil {
+		if err := os.WriteFile(filename, []byte("start %*"), 0644); err != nil {
+			return nil, err
+		}
+	}
+	cmdHelper, err := filepath.Abs(filename)
+	if err != nil {
+		return nil, err
+	}
+
+	if strings.Contains(cmdHelper, " ") {
+		for _, arg := range args {
+			if strings.Contains(arg, " ") {
+				return nil, errors.New("path contains space") // golang bug https://github.com/golang/go/issues/17149#issuecomment-473976818
+			}
+		}
+	}
+	cmd := exec.Command(cmdHelper, args...)
+	cmd.SysProcAttr = &syscall.SysProcAttr{}
+	//go:custom_build windows cmd.SysProcAttr.HideWindow = hideWindow
+	return cmd, nil
+}
+
 func Cmd(args ...string) (string, error) {
 	switch platform := runtime.GOOS; platform {
 	case "windows":
-		if err := os.WriteFile("./cmd-helper.bat", []byte("start %*"), 0644); err != nil {
-			return "", err
-		}
-		cmdHelper, err := filepath.Abs("./cmd-helper")
+		cmd, err := CmdHelper(true, args...)
 		if err != nil {
 			return "", err
 		}
-
-		if strings.Contains(cmdHelper, " ") {
-			for _, arg := range args {
-				if strings.Contains(arg, " ") {
-					return "", errors.New("path contains space") // golang bug https://github.com/golang/go/issues/17149#issuecomment-473976818
-				}
-			}
-		}
-
-		cmd := exec.Command(cmdHelper, args...)
-		out, err := cmd.CombinedOutput()
+		_, err = cmd.CombinedOutput()
 		if err != nil {
 			return "", err
 		}
-		return string(out), nil
+		return "", nil
 	case "darwin":
 		ex, err := os.Executable()
 		if err != nil {

backend-python/convert_pytorch_to_ggml.py

@@ -0,0 +1,169 @@
+# Converts an RWKV model checkpoint in PyTorch format to an rwkv.cpp compatible file.
+# Usage: python convert_pytorch_to_ggml.py C:\RWKV-4-Pile-169M-20220807-8023.pth C:\rwkv.cpp-169M-FP16.bin FP16
+# Get model checkpoints from https://huggingface.co/BlinkDL
+# See FILE_FORMAT.md for the documentation on the file format.
+
+import argparse
+import struct
+import torch
+from typing import Dict
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(
+        description="Convert an RWKV model checkpoint in PyTorch format to an rwkv.cpp compatible file"
+    )
+    parser.add_argument("src_path", help="Path to PyTorch checkpoint file")
+    parser.add_argument(
+        "dest_path", help="Path to rwkv.cpp checkpoint file, will be overwritten"
+    )
+    parser.add_argument(
+        "data_type",
+        help="Data type, FP16, Q4_0, Q4_1, Q5_0, Q5_1, Q8_0",
+        type=str,
+        choices=[
+            "FP16",
+            "Q4_0",
+            "Q4_1",
+            "Q5_0",
+            "Q5_1",
+            "Q8_0",
+        ],
+        default="FP16",
+    )
+    return parser.parse_args()
+
+
+def get_layer_count(state_dict: Dict[str, torch.Tensor]) -> int:
+    n_layer: int = 0
+
+    while f"blocks.{n_layer}.ln1.weight" in state_dict:
+        n_layer += 1
+
+    assert n_layer > 0
+
+    return n_layer
+
+
+def write_state_dict(
+    state_dict: Dict[str, torch.Tensor], dest_path: str, data_type: str
+) -> None:
+    emb_weight: torch.Tensor = state_dict["emb.weight"]
+
+    n_layer: int = get_layer_count(state_dict)
+    n_vocab: int = emb_weight.shape[0]
+    n_embed: int = emb_weight.shape[1]
+
+    is_v5_1_or_2: bool = "blocks.0.att.ln_x.weight" in state_dict
+    is_v5_2: bool = "blocks.0.att.gate.weight" in state_dict
+
+    if is_v5_2:
+        print("Detected RWKV v5.2")
+    elif is_v5_1_or_2:
+        print("Detected RWKV v5.1")
+    else:
+        print("Detected RWKV v4")
+
+    with open(dest_path, "wb") as out_file:
+        is_FP16: bool = data_type == "FP16" or data_type == "float16"
+
+        out_file.write(
+            struct.pack(
+                # Disable padding with '='
+                "=iiiiii",
+                # Magic: 'ggmf' in hex
+                0x67676D66,
+                101,
+                n_vocab,
+                n_embed,
+                n_layer,
+                1 if is_FP16 else 0,
+            )
+        )
+
+        for k in state_dict.keys():
+            tensor: torch.Tensor = state_dict[k].float()
+
+            if ".time_" in k:
+                tensor = tensor.squeeze()
+
+            if is_v5_1_or_2:
+                if ".time_decay" in k:
+                    if is_v5_2:
+                        tensor = torch.exp(-torch.exp(tensor)).unsqueeze(-1)
+                    else:
+                        tensor = torch.exp(-torch.exp(tensor)).reshape(-1, 1, 1)
+
+                if ".time_first" in k:
+                    tensor = torch.exp(tensor).reshape(-1, 1, 1)
+
+                if ".time_faaaa" in k:
+                    tensor = tensor.unsqueeze(-1)
+            else:
+                if ".time_decay" in k:
+                    tensor = -torch.exp(tensor)
+
+            # Keep 1-dim vectors and small matrices in FP32
+            if is_FP16 and len(tensor.shape) > 1 and ".time_" not in k:
+                tensor = tensor.half()
+
+            shape = tensor.shape
+
+            print(f"Writing {k}, shape {shape}, type {tensor.dtype}")
+
+            k_encoded: bytes = k.encode("utf-8")
+
+            out_file.write(
+                struct.pack(
+                    "=iii",
+                    len(shape),
+                    len(k_encoded),
+                    1 if tensor.dtype == torch.float16 else 0,
+                )
+            )
+
+            # Dimension order is reversed here:
+            # * PyTorch shape is (x rows, y columns)
+            # * ggml shape is (y elements in a row, x elements in a column)
+            # Both shapes represent the same tensor.
+            for dim in reversed(tensor.shape):
+                out_file.write(struct.pack("=i", dim))
+
+            out_file.write(k_encoded)
+
+            tensor.numpy().tofile(out_file)
+
+
+def main() -> None:
+    args = parse_args()
+
+    print(f"Reading {args.src_path}")
+
+    state_dict: Dict[str, torch.Tensor] = torch.load(args.src_path, map_location="cpu")
+
+    temp_output: str = args.dest_path
+    if args.data_type.startswith("Q"):
+        import re
+
+        temp_output = re.sub(r"Q[4,5,8]_[0,1]", "fp16", temp_output)
+    write_state_dict(state_dict, temp_output, "FP16")
+    if args.data_type.startswith("Q"):
+        import sys
+        import os
+
+        sys.path.append(os.path.dirname(os.path.realpath(__file__)))
+        from rwkv_pip.cpp import rwkv_cpp_shared_library
+
+        library = rwkv_cpp_shared_library.load_rwkv_shared_library()
+        library.rwkv_quantize_model_file(temp_output, args.dest_path, args.data_type)
+
+    print("Done")
+
+
+if __name__ == "__main__":
+    try:
+        main()
+    except Exception as e:
+        print(e)
+        with open("error.txt", "w") as f:
+            f.write(str(e))

backend-python/convert_safetensors.py

@@ -25,21 +25,50 @@ def rename_key(rename, name):
     return name
 
 
-def convert_file(pt_filename: str, sf_filename: str, transpose_names=[], rename={}):
+def convert_file(pt_filename: str, sf_filename: str, rename={}, transpose_names=[]):
     loaded = torch.load(pt_filename, map_location="cpu")
     if "state_dict" in loaded:
         loaded = loaded["state_dict"]
 
+    kk = list(loaded.keys())
+    version = 4
+    for x in kk:
+        if "ln_x" in x:
+            version = max(5, version)
+        if "gate.weight" in x:
+            version = max(5.1, version)
+        if int(version) == 5 and "att.time_decay" in x:
+            if len(loaded[x].shape) > 1:
+                if loaded[x].shape[1] > 1:
+                    version = max(5.2, version)
+        if "time_maa" in x:
+            version = max(6, version)
+
+    if version == 5.1 and "midi" in pt_filename.lower():
+        import numpy as np
+
+        np.set_printoptions(precision=4, suppress=True, linewidth=200)
+        kk = list(loaded.keys())
+        _, n_emb = loaded["emb.weight"].shape
+        for k in kk:
+            if "time_decay" in k or "time_faaaa" in k:
+                # print(k, mm[k].shape)
+                loaded[k] = (
+                    loaded[k].unsqueeze(1).repeat(1, n_emb // loaded[k].shape[0])
+                )
+
     loaded = {k: v.clone().half() for k, v in loaded.items()}
     # for k, v in loaded.items():
     #     print(f'{k}\t{v.shape}\t{v.dtype}')
 
+    loaded = {rename_key(rename, k).lower(): v.contiguous() for k, v in loaded.items()}
     # For tensors to be contiguous
     for k, v in loaded.items():
         for transpose_name in transpose_names:
             if transpose_name in k:
                 loaded[k] = v.transpose(0, 1)
-    loaded = {rename_key(rename, k).lower(): v.contiguous() for k, v in loaded.items()}
+
+    loaded = {k: v.clone().half().contiguous() for k, v in loaded.items()}
 
     for k, v in loaded.items():
         print(f"{k}\t{v.shape}\t{v.dtype}")
@@ -60,8 +89,18 @@ if __name__ == "__main__":
         convert_file(
             args.input,
             args.output,
-            ["lora_A"],
-            {"time_faaaa": "time_first", "lora_A": "lora.0", "lora_B": "lora.1"},
+            rename={
+                "time_faaaa": "time_first",
+                "time_maa": "time_mix",
+                "lora_A": "lora.0",
+                "lora_B": "lora.1",
+            },
+            transpose_names=[
+                "time_mix_w1",
+                "time_mix_w2",
+                "time_decay_w1",
+                "time_decay_w2",
+            ],
         )
         print(f"Saved to {args.output}")
     except Exception as e:

backend-python/main.py

@@ -32,6 +32,16 @@ def get_args(args: Union[Sequence[str], None] = None):
         action="store_true",
         help="whether to use rwkv-beta (default: False)",
     )
+    group.add_argument(
+        "--rwkv.cpp",
+        action="store_true",
+        help="whether to use rwkv.cpp (default: False)",
+    )
+    group.add_argument(
+        "--webgpu",
+        action="store_true",
+        help="whether to use webgpu (default: False)",
+    )
     args = parser.parse_args(args)
 
     return args

backend-python/requirements.txt

@@ -1,7 +1,7 @@
 torch
 torchvision
 torchaudio
-rwkv==0.8.16
+rwkv==0.8.22
 langchain==0.0.322
 fastapi==0.104.0
 uvicorn==0.23.2

backend-python/requirements_without_cyac.txt

@@ -1,7 +1,7 @@
 torch
 torchvision
 torchaudio
-rwkv==0.8.16
+rwkv==0.8.22
 langchain==0.0.322
 fastapi==0.104.0
 uvicorn==0.23.2

backend-python/routes/completion.py

@@ -8,7 +8,6 @@ import base64
 from fastapi import APIRouter, Request, status, HTTPException
 from sse_starlette.sse import EventSourceResponse
 from pydantic import BaseModel, Field
-import numpy as np
 import tiktoken
 from utils.rwkv import *
 from utils.log import quick_log
@@ -35,6 +34,11 @@ default_stop = [
     "\n\nQ",
     "\n\nHuman",
     "\n\nBob",
+    "\n\nAssistant",
+    "\n\nAnswer",
+    "\n\nA",
+    "\n\nBot",
+    "\n\nAlice",
 ]
 
 
@@ -391,6 +395,8 @@ class EmbeddingsBody(BaseModel):
 
 
 def embedding_base64(embedding: List[float]) -> str:
+    import numpy as np
+
     return base64.b64encode(np.array(embedding).astype(np.float32)).decode("utf-8")
 
 

backend-python/routes/config.py

@@ -25,7 +25,7 @@ class SwitchModelBody(BaseModel):
             "example": {
                 "model": "models/RWKV-4-World-3B-v1-20230619-ctx4096.pth",
                 "strategy": "cuda fp16",
-                "tokenizer": None,
+                "tokenizer": "",
                 "customCuda": False,
                 "deploy": False,
             }
@@ -49,13 +49,20 @@ def switch_model(body: SwitchModelBody, response: Response, request: Request):
     if body.model == "":
         return "success"
 
-    if "->" in body.strategy:
-        state_cache.disable_state_cache()
-    else:
-        try:
-            state_cache.enable_state_cache()
-        except HTTPException:
-            pass
+    devices = set(
+        [
+            x.strip().split(" ")[0].replace("cuda:0", "cuda")
+            for x in body.strategy.split("->")
+        ]
+    )
+    print(f"Strategy Devices: {devices}")
+    # if len(devices) > 1:
+    #     state_cache.disable_state_cache()
+    # else:
+    try:
+        state_cache.enable_state_cache()
+    except HTTPException:
+        pass
 
     os.environ["RWKV_CUDA_ON"] = "1" if body.customCuda else "0"
 
@@ -67,6 +74,10 @@ def switch_model(body: SwitchModelBody, response: Response, request: Request):
         )
     except Exception as e:
         print(e)
+        import traceback
+
+        print(traceback.format_exc())
+
         quick_log(request, body, f"Exception: {e}")
         global_var.set(global_var.Model_Status, global_var.ModelStatus.Offline)
         raise HTTPException(

backend-python/routes/midi.py

@@ -1,6 +1,6 @@
 import io
 import global_var
-from fastapi import APIRouter, HTTPException, status
+from fastapi import APIRouter, HTTPException, UploadFile, status
 from starlette.responses import StreamingResponse
 from pydantic import BaseModel
 from utils.midi import *
@@ -33,6 +33,20 @@ def text_to_midi(body: TextToMidiBody):
     return StreamingResponse(mid_data, media_type="audio/midi")
 
 
+@router.post("/midi-to-text", tags=["MIDI"])
+async def midi_to_text(file_data: UploadFile):
+    vocab_config = "backend-python/utils/midi_vocab_config.json"
+    cfg = VocabConfig.from_json(vocab_config)
+    filter_config = "backend-python/utils/midi_filter_config.json"
+    filter_cfg = FilterConfig.from_json(filter_config)
+    mid = mido.MidiFile(file=file_data.file)
+    output_list = convert_midi_to_str(cfg, filter_cfg, mid)
+    if len(output_list) == 0:
+        raise HTTPException(status.HTTP_400_BAD_REQUEST, "bad midi file")
+
+    return {"text": output_list[0]}
+
+
 class TxtToMidiBody(BaseModel):
     txt_path: str
     midi_path: str

backend-python/routes/state_cache.py

@@ -44,6 +44,7 @@ def disable_state_cache():
     dtrie = {}
     gc.collect()
 
+    print("state cache disabled")
     return "success"
 
 
@@ -61,8 +62,10 @@ def enable_state_cache():
         dtrie = {}
         gc.collect()
 
+        print("state cache enabled")
         return "success"
     except ModuleNotFoundError:
+        print("state cache disabled")
         raise HTTPException(status.HTTP_400_BAD_REQUEST, "cyac not found")
 
 
@@ -73,28 +76,47 @@ class AddStateBody(BaseModel):
     logits: Any
 
 
-@router.post("/add-state", tags=["State Cache"])
+# @router.post("/add-state", tags=["State Cache"])
 def add_state(body: AddStateBody):
     global trie, dtrie, loop_del_trie_id
 
-    if global_var.get(global_var.Deploy_Mode) is True:
-        raise HTTPException(status.HTTP_403_FORBIDDEN)
+    # if global_var.get(global_var.Deploy_Mode) is True:
+    #     raise HTTPException(status.HTTP_403_FORBIDDEN)
 
     if trie is None:
         raise HTTPException(status.HTTP_400_BAD_REQUEST, "trie not loaded")
 
     import torch
+    import numpy as np
 
     try:
+        devices: List[torch.device] = []
+        state: Union[Any, None] = None
+
+        if body.state is not None:
+            if type(body.state) == list or type(body.state) == np.ndarray:
+                devices = [
+                    (
+                        tensor.device
+                        if hasattr(tensor, "device")
+                        else torch.device("cpu")
+                    )
+                    for tensor in body.state
+                ]
+                state = (
+                    [tensor.cpu() for tensor in body.state]
+                    if hasattr(body.state[0], "device")
+                    else copy.deepcopy(body.state)
+                )
+            else:
+                pass  # WebGPU
+
         id: int = trie.insert(body.prompt)
-        device: torch.device = body.state[0].device
         dtrie[id] = {
             "tokens": copy.deepcopy(body.tokens),
-            "state": [tensor.cpu() for tensor in body.state]
-            if device != torch.device("cpu")
-            else copy.deepcopy(body.state),
+            "state": state,
             "logits": copy.deepcopy(body.logits),
-            "device": device,
+            "devices": devices,
         }
 
         if len(trie) >= max_trie_len:
@@ -157,17 +179,18 @@ def __get_a_dtrie_buff_size(dtrie_v):
     return 54 * len(dtrie_v["tokens"]) + 491520 + 262144 + 28  # TODO
 
 
-@router.post("/longest-prefix-state", tags=["State Cache"])
+# @router.post("/longest-prefix-state", tags=["State Cache"])
 def longest_prefix_state(body: LongestPrefixStateBody, request: Request):
     global trie
 
-    if global_var.get(global_var.Deploy_Mode) is True:
-        raise HTTPException(status.HTTP_403_FORBIDDEN)
+    # if global_var.get(global_var.Deploy_Mode) is True:
+    #     raise HTTPException(status.HTTP_403_FORBIDDEN)
 
     if trie is None:
         raise HTTPException(status.HTTP_400_BAD_REQUEST, "trie not loaded")
 
     import torch
+    import numpy as np
 
     id = -1
     try:
@@ -177,35 +200,30 @@ def longest_prefix_state(body: LongestPrefixStateBody, request: Request):
         pass
     if id != -1:
         v = dtrie[id]
-        device: torch.device = v["device"]
+        devices: List[torch.device] = v["devices"]
         prompt: str = trie[id]
+        state: Union[Any, None] = v["state"]
+
+        if state is not None and type(state) == list and hasattr(state[0], "device"):
+            state = [tensor.to(devices[i]) for i, tensor in enumerate(state)]
 
         quick_log(request, body, "Hit:\n" + prompt)
         return {
             "prompt": prompt,
             "tokens": v["tokens"],
-            "state": [tensor.to(device) for tensor in v["state"]]
-            if device != torch.device("cpu")
-            else v["state"],
+            "state": state,
             "logits": v["logits"],
-            "device": device.type,
         }
     else:
-        return {
-            "prompt": "",
-            "tokens": [],
-            "state": None,
-            "logits": None,
-            "device": None,
-        }
+        return {"prompt": "", "tokens": [], "state": None, "logits": None}
 
 
-@router.post("/save-state", tags=["State Cache"])
+# @router.post("/save-state", tags=["State Cache"])
 def save_state():
     global trie
 
-    if global_var.get(global_var.Deploy_Mode) is True:
-        raise HTTPException(status.HTTP_403_FORBIDDEN)
+    # if global_var.get(global_var.Deploy_Mode) is True:
+    #     raise HTTPException(status.HTTP_403_FORBIDDEN)
 
     if trie is None:
         raise HTTPException(status.HTTP_400_BAD_REQUEST, "trie not loaded")

backend-python/rwkv_pip/cpp/model.py

@@ -0,0 +1,14 @@
+from typing import Any, List, Union
+from . import rwkv_cpp_model
+from . import rwkv_cpp_shared_library
+
+
+class RWKV:
+    def __init__(self, model_path: str, strategy=None):
+        self.library = rwkv_cpp_shared_library.load_rwkv_shared_library()
+        self.model = rwkv_cpp_model.RWKVModel(self.library, model_path)
+        self.w = {}  # fake weight
+        self.w["emb.weight"] = [0] * self.model.n_vocab
+
+    def forward(self, tokens: List[int], state: Union[Any, None] = None):
+        return self.model.eval_sequence_in_chunks(tokens, state, use_numpy=True)

backend-python/rwkv_pip/cpp/rwkv_cpp_model.py

@@ -0,0 +1,369 @@
+import os
+import multiprocessing
+
+# Pre-import PyTorch, if available.
+# This fixes "OSError: [WinError 127] The specified procedure could not be found".
+try:
+    import torch
+except ModuleNotFoundError:
+    pass
+
+# I'm sure this is not strictly correct, but let's keep this crutch for now.
+try:
+    import rwkv_cpp_shared_library
+except ModuleNotFoundError:
+    from . import rwkv_cpp_shared_library
+
+from typing import TypeVar, Optional, Tuple, List
+
+# A value of this type is either a numpy's ndarray or a PyTorch's Tensor.
+NumpyArrayOrPyTorchTensor: TypeVar = TypeVar('NumpyArrayOrPyTorchTensor')
+
+class RWKVModel:
+    """
+    An RWKV model managed by rwkv.cpp library.
+    """
+
+    def __init__(
+            self,
+            shared_library: rwkv_cpp_shared_library.RWKVSharedLibrary,
+            model_path: str,
+            thread_count: int = max(1, multiprocessing.cpu_count() // 2),
+            gpu_layer_count: int = 0,
+            **kwargs
+    ) -> None:
+        """
+        Loads the model and prepares it for inference.
+        In case of any error, this method will throw an exception.
+
+        Parameters
+        ----------
+        shared_library : RWKVSharedLibrary
+            rwkv.cpp shared library.
+        model_path : str
+            Path to RWKV model file in ggml format.
+        thread_count : int
+            Thread count to use. If not set, defaults to CPU count / 2.
+        gpu_layer_count : int
+            Count of layers to offload onto the GPU, must be >= 0.
+            See documentation of `gpu_offload_layers` for details about layer offloading.
+        """
+
+        if 'gpu_layers_count' in kwargs:
+            gpu_layer_count = kwargs['gpu_layers_count']
+
+        assert os.path.isfile(model_path), f'{model_path} is not a file'
+        assert thread_count > 0, 'Thread count must be > 0'
+        assert gpu_layer_count >= 0, 'GPU layer count must be >= 0'
+
+        self._library: rwkv_cpp_shared_library.RWKVSharedLibrary = shared_library
+
+        self._ctx: rwkv_cpp_shared_library.RWKVContext = self._library.rwkv_init_from_file(model_path, thread_count)
+
+        if gpu_layer_count > 0:
+            self.gpu_offload_layers(gpu_layer_count)
+
+        self._state_buffer_element_count: int = self._library.rwkv_get_state_buffer_element_count(self._ctx)
+        self._logits_buffer_element_count: int = self._library.rwkv_get_logits_buffer_element_count(self._ctx)
+
+        self._valid: bool = True
+
+    def gpu_offload_layers(self, layer_count: int) -> bool:
+        """
+        Offloads specified count of model layers onto the GPU. Offloaded layers are evaluated using cuBLAS or CLBlast.
+        For the purposes of this function, model head (unembedding matrix) is treated as an additional layer:
+        - pass `model.n_layer` to offload all layers except model head
+        - pass `model.n_layer + 1` to offload all layers, including model head
+
+        Returns true if at least one layer was offloaded.
+        If rwkv.cpp was compiled without cuBLAS and CLBlast support, this function is a no-op and always returns false.
+
+        Parameters
+        ----------
+        layer_count : int
+            Count of layers to offload onto the GPU, must be >= 0.
+        """
+
+        assert layer_count >= 0, 'Layer count must be >= 0'
+
+        return self._library.rwkv_gpu_offload_layers(self._ctx, layer_count)
+
+    @property
+    def n_vocab(self) -> int:
+        return self._library.rwkv_get_n_vocab(self._ctx)
+
+    @property
+    def n_embed(self) -> int:
+        return self._library.rwkv_get_n_embed(self._ctx)
+
+    @property
+    def n_layer(self) -> int:
+        return self._library.rwkv_get_n_layer(self._ctx)
+
+    def eval(
+            self,
+            token: int,
+            state_in: Optional[NumpyArrayOrPyTorchTensor],
+            state_out: Optional[NumpyArrayOrPyTorchTensor] = None,
+            logits_out: Optional[NumpyArrayOrPyTorchTensor] = None,
+            use_numpy: bool = False
+    ) -> Tuple[NumpyArrayOrPyTorchTensor, NumpyArrayOrPyTorchTensor]:
+        """
+        Evaluates the model for a single token.
+        In case of any error, this method will throw an exception.
+
+        Parameters
+        ----------
+        token : int
+            Index of next token to be seen by the model. Must be in range 0 <= token < n_vocab.
+        state_in : Optional[NumpyArrayOrTorchTensor]
+            State from previous call of this method. If this is a first pass, set it to None.
+        state_out : Optional[NumpyArrayOrTorchTensor]
+            Optional output tensor for state. If provided, must be of type float32, contiguous and of shape (state_buffer_element_count).
+        logits_out : Optional[NumpyArrayOrTorchTensor]
+            Optional output tensor for logits. If provided, must be of type float32, contiguous and of shape (logits_buffer_element_count).
+        use_numpy : bool
+            If set to True, numpy's ndarrays will be created instead of PyTorch's Tensors.
+            This parameter is ignored if any tensor parameter is not None; in such case,
+            type of returned tensors will match the type of received tensors.
+
+        Returns
+        -------
+        logits, state
+            Logits vector of shape (n_vocab); state for the next step.
+        """
+
+        assert self._valid, 'Model was freed'
+
+        use_numpy = self._detect_numpy_usage([state_in, state_out, logits_out], use_numpy)
+
+        if state_in is not None:
+            self._validate_tensor(state_in, 'state_in', self._state_buffer_element_count)
+
+            state_in_ptr = self._get_data_ptr(state_in)
+        else:
+            state_in_ptr = 0
+
+        if state_out is not None:
+            self._validate_tensor(state_out, 'state_out', self._state_buffer_element_count)
+        else:
+            state_out = self._zeros_float32(self._state_buffer_element_count, use_numpy)
+
+        if logits_out is not None:
+            self._validate_tensor(logits_out, 'logits_out', self._logits_buffer_element_count)
+        else:
+            logits_out = self._zeros_float32(self._logits_buffer_element_count, use_numpy)
+
+        self._library.rwkv_eval(
+            self._ctx,
+            token,
+            state_in_ptr,
+            self._get_data_ptr(state_out),
+            self._get_data_ptr(logits_out)
+        )
+
+        return logits_out, state_out
+    
+    def eval_sequence(
+            self,
+            tokens: List[int],
+            state_in: Optional[NumpyArrayOrPyTorchTensor],
+            state_out: Optional[NumpyArrayOrPyTorchTensor] = None,
+            logits_out: Optional[NumpyArrayOrPyTorchTensor] = None,
+            use_numpy: bool = False
+    ) -> Tuple[NumpyArrayOrPyTorchTensor, NumpyArrayOrPyTorchTensor]:
+        """
+        Evaluates the model for a sequence of tokens.
+
+        NOTE ON GGML NODE LIMIT
+
+        ggml has a hard-coded limit on max amount of nodes in a computation graph. The sequence graph is built in a way that quickly exceedes
+        this limit when using large models and/or large sequence lengths.
+        Fortunately, rwkv.cpp's fork of ggml has increased limit which was tested to work for sequence lengths up to 64 for 14B models.
+
+        If you get `GGML_ASSERT: ...\\ggml.c:16941: cgraph->n_nodes < GGML_MAX_NODES`, this means you've exceeded the limit.
+        To get rid of the assertion failure, reduce the model size and/or sequence length.
+
+        In case of any error, this method will throw an exception.
+
+        Parameters
+        ----------
+        tokens : List[int]
+            Indices of the next tokens to be seen by the model. Must be in range 0 <= token < n_vocab.
+        state_in : Optional[NumpyArrayOrTorchTensor]
+            State from previous call of this method. If this is a first pass, set it to None.
+        state_out : Optional[NumpyArrayOrTorchTensor]
+            Optional output tensor for state. If provided, must be of type float32, contiguous and of shape (state_buffer_element_count).
+        logits_out : Optional[NumpyArrayOrTorchTensor]
+            Optional output tensor for logits. If provided, must be of type float32, contiguous and of shape (logits_buffer_element_count).
+        use_numpy : bool
+            If set to True, numpy's ndarrays will be created instead of PyTorch's Tensors.
+            This parameter is ignored if any tensor parameter is not None; in such case,
+            type of returned tensors will match the type of received tensors.
+
+        Returns
+        -------
+        logits, state
+            Logits vector of shape (n_vocab); state for the next step.
+        """
+
+        assert self._valid, 'Model was freed'
+
+        use_numpy = self._detect_numpy_usage([state_in, state_out, logits_out], use_numpy)
+
+        if state_in is not None:
+            self._validate_tensor(state_in, 'state_in', self._state_buffer_element_count)
+
+            state_in_ptr = self._get_data_ptr(state_in)
+        else:
+            state_in_ptr = 0
+
+        if state_out is not None:
+            self._validate_tensor(state_out, 'state_out', self._state_buffer_element_count)
+        else:
+            state_out = self._zeros_float32(self._state_buffer_element_count, use_numpy)
+
+        if logits_out is not None:
+            self._validate_tensor(logits_out, 'logits_out', self._logits_buffer_element_count)
+        else:
+            logits_out = self._zeros_float32(self._logits_buffer_element_count, use_numpy)
+
+        self._library.rwkv_eval_sequence(
+            self._ctx,
+            tokens,
+            state_in_ptr,
+            self._get_data_ptr(state_out),
+            self._get_data_ptr(logits_out)
+        )
+
+        return logits_out, state_out
+
+    def eval_sequence_in_chunks(
+            self,
+            tokens: List[int],
+            state_in: Optional[NumpyArrayOrPyTorchTensor],
+            state_out: Optional[NumpyArrayOrPyTorchTensor] = None,
+            logits_out: Optional[NumpyArrayOrPyTorchTensor] = None,
+            chunk_size: int = 16,
+            use_numpy: bool = False
+    ) -> Tuple[NumpyArrayOrPyTorchTensor, NumpyArrayOrPyTorchTensor]:
+        """
+        Evaluates the model for a sequence of tokens using `eval_sequence`, splitting a potentially long sequence into fixed-length chunks.
+        This function is useful for processing complete prompts and user input in chat & role-playing use-cases.
+        It is recommended to use this function instead of `eval_sequence` to avoid mistakes and get maximum performance.
+
+        Chunking allows processing sequences of thousands of tokens, while not reaching the ggml's node limit and not consuming too much memory.
+        A reasonable and recommended value of chunk size is 16. If you want maximum performance, try different chunk sizes in range [2..64]
+        and choose one that works the best in your use case.
+
+        In case of any error, this method will throw an exception.
+
+        Parameters
+        ----------
+        tokens : List[int]
+            Indices of the next tokens to be seen by the model. Must be in range 0 <= token < n_vocab.
+        chunk_size : int
+            Size of each chunk in tokens, must be positive.
+        state_in : Optional[NumpyArrayOrTorchTensor]
+            State from previous call of this method. If this is a first pass, set it to None.
+        state_out : Optional[NumpyArrayOrTorchTensor]
+            Optional output tensor for state. If provided, must be of type float32, contiguous and of shape (state_buffer_element_count).
+        logits_out : Optional[NumpyArrayOrTorchTensor]
+            Optional output tensor for logits. If provided, must be of type float32, contiguous and of shape (logits_buffer_element_count).
+        use_numpy : bool
+            If set to True, numpy's ndarrays will be created instead of PyTorch's Tensors.
+            This parameter is ignored if any tensor parameter is not None; in such case,
+            type of returned tensors will match the type of received tensors.
+
+        Returns
+        -------
+        logits, state
+            Logits vector of shape (n_vocab); state for the next step.
+        """
+
+        assert self._valid, 'Model was freed'
+
+        use_numpy = self._detect_numpy_usage([state_in, state_out, logits_out], use_numpy)
+
+        if state_in is not None:
+            self._validate_tensor(state_in, 'state_in', self._state_buffer_element_count)
+
+            state_in_ptr = self._get_data_ptr(state_in)
+        else:
+            state_in_ptr = 0
+
+        if state_out is not None:
+            self._validate_tensor(state_out, 'state_out', self._state_buffer_element_count)
+        else:
+            state_out = self._zeros_float32(self._state_buffer_element_count, use_numpy)
+
+        if logits_out is not None:
+            self._validate_tensor(logits_out, 'logits_out', self._logits_buffer_element_count)
+        else:
+            logits_out = self._zeros_float32(self._logits_buffer_element_count, use_numpy)
+
+        self._library.rwkv_eval_sequence_in_chunks(
+            self._ctx,
+            tokens,
+            chunk_size,
+            state_in_ptr,
+            self._get_data_ptr(state_out),
+            self._get_data_ptr(logits_out)
+        )
+
+        return logits_out, state_out
+
+    def free(self) -> None:
+        """
+        Frees all allocated resources.
+        In case of any error, this method will throw an exception.
+        The object must not be used anymore after calling this method.
+        """
+
+        assert self._valid, 'Already freed'
+
+        self._valid = False
+
+        self._library.rwkv_free(self._ctx)
+
+    def __del__(self) -> None:
+        # Free the context on GC in case user forgot to call free() explicitly.
+        if hasattr(self, '_valid') and self._valid:
+            self.free()
+
+    def _is_pytorch_tensor(self, tensor: NumpyArrayOrPyTorchTensor) -> bool:
+        return hasattr(tensor, '__module__') and tensor.__module__ == 'torch'
+
+    def _detect_numpy_usage(self, tensors: List[Optional[NumpyArrayOrPyTorchTensor]], use_numpy_by_default: bool) -> bool:
+        for tensor in tensors:
+            if tensor is not None:
+                return False if self._is_pytorch_tensor(tensor) else True
+
+        return use_numpy_by_default
+
+    def _validate_tensor(self, tensor: NumpyArrayOrPyTorchTensor, name: str, size: int) -> None:
+        if self._is_pytorch_tensor(tensor):
+            tensor: torch.Tensor = tensor
+            assert tensor.device == torch.device('cpu'), f'{name} is not on CPU'
+            assert tensor.dtype == torch.float32, f'{name} is not of type float32'
+            assert tensor.shape == (size,), f'{name} has invalid shape {tensor.shape}, expected ({size})'
+            assert tensor.is_contiguous(), f'{name} is not contiguous'
+        else:
+            import numpy as np
+            tensor: np.ndarray = tensor
+            assert tensor.dtype == np.float32, f'{name} is not of type float32'
+            assert tensor.shape == (size,), f'{name} has invalid shape {tensor.shape}, expected ({size})'
+            assert tensor.data.contiguous, f'{name} is not contiguous'
+
+    def _get_data_ptr(self, tensor: NumpyArrayOrPyTorchTensor):
+        if self._is_pytorch_tensor(tensor):
+            return tensor.data_ptr()
+        else:
+            return tensor.ctypes.data
+
+    def _zeros_float32(self, element_count: int, use_numpy: bool) -> NumpyArrayOrPyTorchTensor:
+        if use_numpy:
+            import numpy as np
+            return np.zeros(element_count, dtype=np.float32)
+        else:
+            return torch.zeros(element_count, dtype=torch.float32, device='cpu')

backend-python/rwkv_pip/cpp/rwkv_cpp_shared_library.py

@@ -0,0 +1,444 @@
+import os
+import sys
+import ctypes
+import pathlib
+import platform
+from typing import Optional, List, Tuple, Callable
+
+QUANTIZED_FORMAT_NAMES: Tuple[str, str, str, str, str] = (
+    'Q4_0',
+    'Q4_1',
+    'Q5_0',
+    'Q5_1',
+    'Q8_0'
+)
+
+P_FLOAT = ctypes.POINTER(ctypes.c_float)
+P_INT = ctypes.POINTER(ctypes.c_int32)
+
+class RWKVContext:
+
+    def __init__(self, ptr: ctypes.pointer) -> None:
+        self.ptr: ctypes.pointer = ptr
+
+class RWKVSharedLibrary:
+    """
+    Python wrapper around rwkv.cpp shared library.
+    """
+
+    def __init__(self, shared_library_path: str) -> None:
+        """
+        Loads the shared library from specified file.
+        In case of any error, this method will throw an exception.
+
+        Parameters
+        ----------
+        shared_library_path : str
+            Path to rwkv.cpp shared library. On Windows, it would look like 'rwkv.dll'. On UNIX, 'rwkv.so'.
+        """
+        #  When Python is greater than 3.8, we need to reprocess the custom dll
+        #  according to the documentation to prevent loading failure errors.
+        #  https://docs.python.org/3/whatsnew/3.8.html#ctypes
+        if platform.system().lower() == 'windows':
+            self.library = ctypes.CDLL(shared_library_path, winmode=0)
+        else:
+            self.library = ctypes.cdll.LoadLibrary(shared_library_path)
+
+        self.library.rwkv_init_from_file.argtypes = [ctypes.c_char_p, ctypes.c_uint32]
+        self.library.rwkv_init_from_file.restype = ctypes.c_void_p
+
+        self.library.rwkv_gpu_offload_layers.argtypes = [ctypes.c_void_p, ctypes.c_uint32]
+        self.library.rwkv_gpu_offload_layers.restype = ctypes.c_bool
+
+        self.library.rwkv_eval.argtypes = [
+            ctypes.c_void_p, # ctx
+            ctypes.c_int32, # token
+            P_FLOAT, # state_in
+            P_FLOAT, # state_out
+            P_FLOAT  # logits_out
+        ]
+        self.library.rwkv_eval.restype = ctypes.c_bool
+
+        self.library.rwkv_eval_sequence.argtypes = [
+            ctypes.c_void_p, # ctx
+            P_INT, # tokens
+            ctypes.c_size_t, # token count
+            P_FLOAT, # state_in
+            P_FLOAT, # state_out
+            P_FLOAT  # logits_out
+        ]
+        self.library.rwkv_eval_sequence.restype = ctypes.c_bool
+
+        self.library.rwkv_eval_sequence_in_chunks.argtypes = [
+            ctypes.c_void_p, # ctx
+            P_INT, # tokens
+            ctypes.c_size_t, # token count
+            ctypes.c_size_t, # chunk size
+            P_FLOAT, # state_in
+            P_FLOAT, # state_out
+            P_FLOAT  # logits_out
+        ]
+        self.library.rwkv_eval_sequence_in_chunks.restype = ctypes.c_bool
+
+        self.library.rwkv_get_n_vocab.argtypes = [ctypes.c_void_p]
+        self.library.rwkv_get_n_vocab.restype = ctypes.c_size_t
+
+        self.library.rwkv_get_n_embed.argtypes = [ctypes.c_void_p]
+        self.library.rwkv_get_n_embed.restype = ctypes.c_size_t
+
+        self.library.rwkv_get_n_layer.argtypes = [ctypes.c_void_p]
+        self.library.rwkv_get_n_layer.restype = ctypes.c_size_t
+
+        self.library.rwkv_get_state_buffer_element_count.argtypes = [ctypes.c_void_p]
+        self.library.rwkv_get_state_buffer_element_count.restype = ctypes.c_uint32
+
+        self.library.rwkv_get_logits_buffer_element_count.argtypes = [ctypes.c_void_p]
+        self.library.rwkv_get_logits_buffer_element_count.restype = ctypes.c_uint32
+
+        self.library.rwkv_free.argtypes = [ctypes.c_void_p]
+        self.library.rwkv_free.restype = None
+
+        self.library.rwkv_free.argtypes = [ctypes.c_void_p]
+        self.library.rwkv_free.restype = None
+
+        self.library.rwkv_quantize_model_file.argtypes = [ctypes.c_char_p, ctypes.c_char_p, ctypes.c_char_p]
+        self.library.rwkv_quantize_model_file.restype = ctypes.c_bool
+
+        self.library.rwkv_get_system_info_string.argtypes = []
+        self.library.rwkv_get_system_info_string.restype = ctypes.c_char_p
+
+        self.nullptr = ctypes.cast(0, ctypes.c_void_p)
+
+    def rwkv_init_from_file(self, model_file_path: str, thread_count: int) -> RWKVContext:
+        """
+        Loads the model from a file and prepares it for inference.
+        Throws an exception in case of any error. Error messages would be printed to stderr.
+
+        Parameters
+        ----------
+        model_file_path : str
+            Path to model file in ggml format.
+        thread_count : int
+            Count of threads to use, must be positive.
+        """
+
+        ptr = self.library.rwkv_init_from_file(model_file_path.encode('utf-8'), ctypes.c_uint32(thread_count))
+
+        assert ptr is not None, 'rwkv_init_from_file failed, check stderr'
+
+        return RWKVContext(ptr)
+
+    def rwkv_gpu_offload_layers(self, ctx: RWKVContext, layer_count: int) -> bool:
+        """
+        Offloads specified count of model layers onto the GPU. Offloaded layers are evaluated using cuBLAS or CLBlast.
+        For the purposes of this function, model head (unembedding matrix) is treated as an additional layer:
+        - pass `rwkv_get_n_layer(ctx)` to offload all layers except model head
+        - pass `rwkv_get_n_layer(ctx) + 1` to offload all layers, including model head
+        Returns true if at least one layer was offloaded.
+        If rwkv.cpp was compiled without cuBLAS and CLBlast support, this function is a no-op and always returns false.
+
+        Parameters
+        ----------
+        ctx : RWKVContext
+            RWKV context obtained from rwkv_init_from_file.
+        layer_count : int
+            Count of layers to offload onto the GPU, must be >= 0.
+        """
+
+        assert layer_count >= 0, 'Layer count must be >= 0'
+
+        return self.library.rwkv_gpu_offload_layers(ctx.ptr, ctypes.c_uint32(layer_count))
+
+    def rwkv_eval(
+            self,
+            ctx: RWKVContext,
+            token: int,
+            state_in_address: Optional[int],
+            state_out_address: int,
+            logits_out_address: int
+    ) -> None:
+        """
+        Evaluates the model for a single token.
+        Throws an exception in case of any error. Error messages would be printed to stderr.
+        Not thread-safe. For parallel inference, call rwkv_clone_context to create one rwkv_context for each thread.
+
+        Parameters
+        ----------
+        ctx : RWKVContext
+            RWKV context obtained from rwkv_init_from_file.
+        token : int
+            Next token index, in range 0 <= token < n_vocab.
+        state_in_address : int
+            Address of the first element of a FP32 buffer of size rwkv_get_state_buffer_element_count; or None, if this is a first pass.
+        state_out_address : int
+            Address of the first element of a FP32 buffer of size rwkv_get_state_buffer_element_count. This buffer will be written to.
+        logits_out_address : int
+            Address of the first element of a FP32 buffer of size rwkv_get_logits_buffer_element_count. This buffer will be written to.
+        """
+
+        assert self.library.rwkv_eval(
+            ctx.ptr,
+            ctypes.c_int32(token),
+            ctypes.cast(0 if state_in_address is None else state_in_address, P_FLOAT),
+            ctypes.cast(state_out_address, P_FLOAT),
+            ctypes.cast(logits_out_address, P_FLOAT)
+        ), 'rwkv_eval failed, check stderr'
+
+    def rwkv_eval_sequence(
+            self,
+            ctx: RWKVContext,
+            tokens: List[int],
+            state_in_address: Optional[int],
+            state_out_address: int,
+            logits_out_address: int
+    ) -> None:
+        """
+        Evaluates the model for a sequence of tokens.
+        Uses a faster algorithm than `rwkv_eval` if you do not need the state and logits for every token. Best used with sequence lengths of 64 or so.
+        Has to build a computation graph on the first call for a given sequence, but will use this cached graph for subsequent calls of the same sequence length.
+
+        NOTE ON GGML NODE LIMIT
+
+        ggml has a hard-coded limit on max amount of nodes in a computation graph. The sequence graph is built in a way that quickly exceedes
+        this limit when using large models and/or large sequence lengths.
+        Fortunately, rwkv.cpp's fork of ggml has increased limit which was tested to work for sequence lengths up to 64 for 14B models.
+
+        If you get `GGML_ASSERT: ...\\ggml.c:16941: cgraph->n_nodes < GGML_MAX_NODES`, this means you've exceeded the limit.
+        To get rid of the assertion failure, reduce the model size and/or sequence length.
+
+        Not thread-safe. For parallel inference, call `rwkv_clone_context` to create one rwkv_context for each thread.
+        Throws an exception in case of any error. Error messages would be printed to stderr.
+
+        Parameters
+        ----------
+        ctx : RWKVContext
+            RWKV context obtained from rwkv_init_from_file.
+        tokens : List[int]
+            Next token indices, in range 0 <= token < n_vocab.
+        state_in_address : int
+            Address of the first element of a FP32 buffer of size rwkv_get_state_buffer_element_count; or None, if this is a first pass.
+        state_out_address : int
+            Address of the first element of a FP32 buffer of size rwkv_get_state_buffer_element_count. This buffer will be written to.
+        logits_out_address : int
+            Address of the first element of a FP32 buffer of size rwkv_get_logits_buffer_element_count. This buffer will be written to.
+        """
+
+        assert self.library.rwkv_eval_sequence(
+            ctx.ptr,
+            ctypes.cast((ctypes.c_int32 * len(tokens))(*tokens), P_INT),
+            ctypes.c_size_t(len(tokens)),
+            ctypes.cast(0 if state_in_address is None else state_in_address, P_FLOAT),
+            ctypes.cast(state_out_address, P_FLOAT),
+            ctypes.cast(logits_out_address, P_FLOAT)
+        ), 'rwkv_eval_sequence failed, check stderr'
+
+    def rwkv_eval_sequence_in_chunks(
+            self,
+            ctx: RWKVContext,
+            tokens: List[int],
+            chunk_size: int,
+            state_in_address: Optional[int],
+            state_out_address: int,
+            logits_out_address: int
+    ) -> None:
+        """
+        Evaluates the model for a sequence of tokens using `rwkv_eval_sequence`, splitting a potentially long sequence into fixed-length chunks.
+        This function is useful for processing complete prompts and user input in chat & role-playing use-cases.
+        It is recommended to use this function instead of `rwkv_eval_sequence` to avoid mistakes and get maximum performance.
+
+        Chunking allows processing sequences of thousands of tokens, while not reaching the ggml's node limit and not consuming too much memory.
+        A reasonable and recommended value of chunk size is 16. If you want maximum performance, try different chunk sizes in range [2..64]
+        and choose one that works the best in your use case.
+
+        Not thread-safe. For parallel inference, call `rwkv_clone_context` to create one rwkv_context for each thread.
+        Throws an exception in case of any error. Error messages would be printed to stderr.
+
+        Parameters
+        ----------
+        ctx : RWKVContext
+            RWKV context obtained from rwkv_init_from_file.
+        tokens : List[int]
+            Next token indices, in range 0 <= token < n_vocab.
+        chunk_size : int
+            Size of each chunk in tokens, must be positive.
+        state_in_address : int
+            Address of the first element of a FP32 buffer of size rwkv_get_state_buffer_element_count; or None, if this is a first pass.
+        state_out_address : int
+            Address of the first element of a FP32 buffer of size rwkv_get_state_buffer_element_count. This buffer will be written to.
+        logits_out_address : int
+            Address of the first element of a FP32 buffer of size rwkv_get_logits_buffer_element_count. This buffer will be written to.
+        """
+
+        assert self.library.rwkv_eval_sequence_in_chunks(
+            ctx.ptr,
+            ctypes.cast((ctypes.c_int32 * len(tokens))(*tokens), P_INT),
+            ctypes.c_size_t(len(tokens)),
+            ctypes.c_size_t(chunk_size),
+            ctypes.cast(0 if state_in_address is None else state_in_address, P_FLOAT),
+            ctypes.cast(state_out_address, P_FLOAT),
+            ctypes.cast(logits_out_address, P_FLOAT)
+        ), 'rwkv_eval_sequence_in_chunks failed, check stderr'
+
+    def rwkv_get_n_vocab(self, ctx: RWKVContext) -> int:
+        """
+        Returns the number of tokens in the given model's vocabulary.
+        Useful for telling 20B_tokenizer models (n_vocab = 50277) apart from World models (n_vocab = 65536).
+
+        Parameters
+        ----------
+        ctx : RWKVContext
+            RWKV context obtained from rwkv_init_from_file.
+        """
+
+        return self.library.rwkv_get_n_vocab(ctx.ptr)
+
+    def rwkv_get_n_embed(self, ctx: RWKVContext) -> int:
+        """
+        Returns the number of elements in the given model's embedding.
+        Useful for reading individual fields of a model's hidden state.
+
+        Parameters
+        ----------
+        ctx : RWKVContext
+            RWKV context obtained from rwkv_init_from_file.
+        """
+
+        return self.library.rwkv_get_n_embed(ctx.ptr)
+
+    def rwkv_get_n_layer(self, ctx: RWKVContext) -> int:
+        """
+        Returns the number of layers in the given model.
+        A layer is a pair of RWKV and FFN operations, stacked multiple times throughout the model.
+        Embedding matrix and model head (unembedding matrix) are NOT counted in `n_layer`.
+        Useful for always offloading the entire model to GPU.
+
+        Parameters
+        ----------
+        ctx : RWKVContext
+            RWKV context obtained from rwkv_init_from_file.
+        """
+
+        return self.library.rwkv_get_n_layer(ctx.ptr)
+
+    def rwkv_get_state_buffer_element_count(self, ctx: RWKVContext) -> int:
+        """
+        Returns count of FP32 elements in state buffer.
+
+        Parameters
+        ----------
+        ctx : RWKVContext
+            RWKV context obtained from rwkv_init_from_file.
+        """
+
+        return self.library.rwkv_get_state_buffer_element_count(ctx.ptr)
+
+    def rwkv_get_logits_buffer_element_count(self, ctx: RWKVContext) -> int:
+        """
+        Returns count of FP32 elements in logits buffer.
+
+        Parameters
+        ----------
+        ctx : RWKVContext
+            RWKV context obtained from rwkv_init_from_file.
+        """
+
+        return self.library.rwkv_get_logits_buffer_element_count(ctx.ptr)
+
+    def rwkv_free(self, ctx: RWKVContext) -> None:
+        """
+        Frees all allocated memory and the context.
+
+        Parameters
+        ----------
+        ctx : RWKVContext
+            RWKV context obtained from rwkv_init_from_file.
+        """
+
+        self.library.rwkv_free(ctx.ptr)
+
+        ctx.ptr = self.nullptr
+
+    def rwkv_quantize_model_file(self, model_file_path_in: str, model_file_path_out: str, format_name: str) -> None:
+        """
+        Quantizes FP32 or FP16 model to one of INT4 formats.
+        Throws an exception in case of any error. Error messages would be printed to stderr.
+
+        Parameters
+        ----------
+        model_file_path_in : str
+            Path to model file in ggml format, must be either FP32 or FP16.
+        model_file_path_out : str
+            Quantized model will be written here.
+        format_name : str
+            One of QUANTIZED_FORMAT_NAMES.
+        """
+
+        assert format_name in QUANTIZED_FORMAT_NAMES, f'Unknown format name {format_name}, use one of {QUANTIZED_FORMAT_NAMES}'
+
+        assert self.library.rwkv_quantize_model_file(
+            model_file_path_in.encode('utf-8'),
+            model_file_path_out.encode('utf-8'),
+            format_name.encode('utf-8')
+        ), 'rwkv_quantize_model_file failed, check stderr'
+
+    def rwkv_get_system_info_string(self) -> str:
+        """
+        Returns system information string.
+        """
+
+        return self.library.rwkv_get_system_info_string().decode('utf-8')
+
+def load_rwkv_shared_library() -> RWKVSharedLibrary:
+    """
+    Attempts to find rwkv.cpp shared library and load it.
+    To specify exact path to the library, create an instance of RWKVSharedLibrary explicitly.
+    """
+
+    file_name: str
+
+    if 'win32' in sys.platform or 'cygwin' in sys.platform:
+        file_name = 'rwkv.dll'
+    elif 'darwin' in sys.platform:
+        file_name = 'librwkv.dylib'
+    else:
+        file_name = 'librwkv.so'
+
+    # Possible sub-paths to the library relative to the repo dir.
+    child_paths: List[Callable[[pathlib.Path], pathlib.Path]] = [
+        # No lookup for Debug config here.
+        # I assume that if a user wants to debug the library,
+        # they will be able to find the library and set the exact path explicitly.
+        lambda p: p / 'backend-python' / 'rwkv_pip' / 'cpp' / file_name,
+        lambda p: p / 'bin' / 'Release' / file_name,
+        lambda p: p / 'bin' / file_name,
+        # Some people prefer to build in the "build" subdirectory.
+        lambda p: p / 'build' / 'bin' / 'Release' / file_name,
+        lambda p: p / 'build' / 'bin' / file_name,
+        lambda p: p / 'build' / file_name,
+        # Fallback.
+        lambda p: p / file_name
+    ]
+
+    working_dir: pathlib.Path = pathlib.Path(os.path.abspath(os.getcwd()))
+
+    parent_paths: List[pathlib.Path] = [
+        # Possible repo dirs relative to the working dir.
+        # ./python/rwkv_cpp
+        working_dir.parent.parent,
+        # ./python
+        working_dir.parent,
+        # .
+        working_dir,
+        # Repo dir relative to this Python file.
+        pathlib.Path(os.path.abspath(__file__)).parent.parent.parent
+    ]
+
+    for parent_path in parent_paths:
+        for child_path in child_paths:
+            full_path: pathlib.Path = child_path(parent_path)
+
+            if os.path.isfile(full_path):
+                return RWKVSharedLibrary(str(full_path))
+
+    assert False, (f'Failed to find {file_name} automatically; '
+                   f'you need to find the library and create RWKVSharedLibrary specifying the path to it')

backend-python/rwkv_pip/cuda/rwkv6.cu

@@ -0,0 +1,87 @@
+#include <stdio.h>
+#include <assert.h>
+#include "ATen/ATen.h"
+typedef at::BFloat16 bf16;
+typedef at::Half fp16;
+typedef float fp32;
+
+template <typename F>
+__global__ void kernel_forward(const int B, const int T, const int C, const int H, float *__restrict__ _state,
+                               const F *__restrict__ const _r, const F *__restrict__ const _k, const F *__restrict__ const _v, const float *__restrict__ _w, const F *__restrict__ _u,
+                               F *__restrict__ const _y)
+{
+    const int b = blockIdx.x / H;
+    const int h = blockIdx.x % H;
+    const int i = threadIdx.x;
+    _u += h*_N_;
+    _state += h*_N_*_N_ + i*_N_; // wrong if B > 1 !!!
+
+    __shared__ float r[_N_], k[_N_], u[_N_], w[_N_];
+    
+    float state[_N_];
+    #pragma unroll
+    for (int j = 0; j < _N_; j++)
+        state[j] = _state[j];
+
+    __syncthreads();
+    u[i] = float(_u[i]);
+    __syncthreads();
+
+    for (int t = b*T*C + h*_N_ + i; t < (b+1)*T*C + h*_N_ + i; t += C)
+    {
+        __syncthreads();
+        w[i] = _w[t];
+        r[i] = float(_r[t]);
+        k[i] = float(_k[t]);
+        __syncthreads();
+
+        const float v = float(_v[t]);
+        float y = 0;
+
+        #pragma unroll
+        for (int j = 0; j < _N_; j+=4)
+        {
+            const float4& r_ = (float4&)(r[j]);
+            const float4& k_ = (float4&)(k[j]);
+            const float4& w_ = (float4&)(w[j]);
+            const float4& u_ = (float4&)(u[j]);
+            float4& s = (float4&)(state[j]);
+            float4 x;
+
+            x.x = k_.x * v;
+            x.y = k_.y * v;
+            x.z = k_.z * v;
+            x.w = k_.w * v;
+
+            y += r_.x * (u_.x * x.x + s.x);
+            y += r_.y * (u_.y * x.y + s.y);
+            y += r_.z * (u_.z * x.z + s.z);
+            y += r_.w * (u_.w * x.w + s.w);
+
+            s.x = s.x * w_.x + x.x;
+            s.y = s.y * w_.y + x.y;
+            s.z = s.z * w_.z + x.z;
+            s.w = s.w * w_.w + x.w;
+        }
+        _y[t] = F(y);
+    }
+    #pragma unroll
+    for (int j = 0; j < _N_; j++)
+        _state[j] = state[j];
+}
+
+void cuda_forward_bf16(int B, int T, int C, int H, float *state, bf16 *r, bf16 *k, bf16 *v, float *w, bf16 *u, bf16 *y)
+{
+    assert(H*_N_ == C);
+    kernel_forward<<<dim3(B * H), dim3(_N_)>>>(B, T, C, H, state, r, k, v, w, u, y);
+}
+void cuda_forward_fp16(int B, int T, int C, int H, float *state, fp16 *r, fp16 *k, fp16 *v, float *w, fp16 *u, fp16 *y)
+{
+    assert(H*_N_ == C);
+    kernel_forward<<<dim3(B * H), dim3(_N_)>>>(B, T, C, H, state, r, k, v, w, u, y);
+}
+void cuda_forward_fp32(int B, int T, int C, int H, float *state, fp32 *r, fp32 *k, fp32 *v, float *w, fp32 *u, fp32 *y)
+{
+    assert(H*_N_ == C);
+    kernel_forward<<<dim3(B * H), dim3(_N_)>>>(B, T, C, H, state, r, k, v, w, u, y);
+}

backend-python/rwkv_pip/cuda/rwkv6_op.cpp

@@ -0,0 +1,34 @@
+#include <torch/extension.h>
+#include "ATen/ATen.h"
+#include <c10/cuda/CUDAGuard.h>
+typedef at::BFloat16 bf16;
+typedef at::Half fp16;
+typedef float fp32;
+
+void cuda_forward_bf16(int B, int T, int C, int H, float *state, bf16 *r, bf16 *k, bf16 *v, float *w, bf16 *u, bf16 *y);
+void cuda_forward_fp16(int B, int T, int C, int H, float *state, fp16 *r, fp16 *k, fp16 *v, float *w, fp16 *u, fp16 *y);
+void cuda_forward_fp32(int B, int T, int C, int H, float *state, fp32 *r, fp32 *k, fp32 *v, float *w, fp32 *u, fp32 *y);
+
+void forward_bf16(int64_t B, int64_t T, int64_t C, int64_t H, torch::Tensor &state, torch::Tensor &r, torch::Tensor &k, torch::Tensor &v, torch::Tensor &w, torch::Tensor &u, torch::Tensor &y) {
+    const at::cuda::OptionalCUDAGuard device_guard(device_of(state));
+    cuda_forward_bf16(B, T, C, H, state.data_ptr<float>(), r.data_ptr<bf16>(), k.data_ptr<bf16>(), v.data_ptr<bf16>(), w.data_ptr<float>(), u.data_ptr<bf16>(), y.data_ptr<bf16>());
+}
+void forward_fp16(int64_t B, int64_t T, int64_t C, int64_t H, torch::Tensor &state, torch::Tensor &r, torch::Tensor &k, torch::Tensor &v, torch::Tensor &w, torch::Tensor &u, torch::Tensor &y) {
+    const at::cuda::OptionalCUDAGuard device_guard(device_of(state));
+    cuda_forward_fp16(B, T, C, H, state.data_ptr<float>(), r.data_ptr<fp16>(), k.data_ptr<fp16>(), v.data_ptr<fp16>(), w.data_ptr<float>(), u.data_ptr<fp16>(), y.data_ptr<fp16>());
+}
+void forward_fp32(int64_t B, int64_t T, int64_t C, int64_t H, torch::Tensor &state, torch::Tensor &r, torch::Tensor &k, torch::Tensor &v, torch::Tensor &w, torch::Tensor &u, torch::Tensor &y) {
+    const at::cuda::OptionalCUDAGuard device_guard(device_of(state));
+    cuda_forward_fp32(B, T, C, H, state.data_ptr<float>(), r.data_ptr<fp32>(), k.data_ptr<fp32>(), v.data_ptr<fp32>(), w.data_ptr<float>(), u.data_ptr<fp32>(), y.data_ptr<fp32>());
+}
+
+PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
+    m.def("forward_bf16", &forward_bf16, "rwkv6 forward_bf16");
+    m.def("forward_fp16", &forward_fp16, "rwkv6 forward_fp16");
+    m.def("forward_fp32", &forward_fp32, "rwkv6 forward_fp32");
+}
+TORCH_LIBRARY(rwkv6, m) {
+    m.def("forward_bf16", forward_bf16);
+    m.def("forward_fp16", forward_fp16);
+    m.def("forward_fp32", forward_fp32);
+}

backend-python/rwkv_pip/utils.py

@@ -78,12 +78,24 @@ class PIPELINE:
     def decode(self, x):
         return self.tokenizer.decode(x)
 
+    def np_softmax(self, x: np.ndarray, axis: int):
+        x -= x.max(axis=axis, keepdims=True)
+        e: np.ndarray = np.exp(x)
+        return e / e.sum(axis=axis, keepdims=True)
+
     def sample_logits(self, logits, temperature=1.0, top_p=0.85, top_k=0):
-        probs = F.softmax(logits.float(), dim=-1)
+        if type(logits) == list:
+            logits = np.array(logits)
+        np_logits = type(logits) == np.ndarray
+        if np_logits:
+            probs = self.np_softmax(logits, axis=-1)
+        else:
+            probs = F.softmax(logits.float(), dim=-1)
         top_k = int(top_k)
         # 'privateuseone' is the type of custom devices like `torch_directml.device()`
-        if probs.device.type in ["cpu", "privateuseone"]:
-            probs = probs.cpu().numpy()
+        if np_logits or probs.device.type in ["cpu", "privateuseone"]:
+            if not np_logits:
+                probs = probs.cpu().numpy()
             sorted_ids = np.argsort(probs)
             sorted_probs = probs[sorted_ids][::-1]
             cumulative_probs = np.cumsum(sorted_probs)

backend-python/rwkv_pip/webgpu/model.py

@@ -0,0 +1,26 @@
+from typing import Any, List, Union
+
+try:
+    import web_rwkv_py as wrp
+except ModuleNotFoundError:
+    try:
+        from . import web_rwkv_py as wrp
+    except ImportError:
+        raise ModuleNotFoundError(
+            "web_rwkv_py not found, install it from https://github.com/cryscan/web-rwkv-py"
+        )
+
+
+class RWKV:
+    def __init__(self, model_path: str, strategy: str = None):
+        self.model = wrp.v5.Model(
+            model_path,
+            turbo=False,
+            quant=32 if "i8" in strategy else None,
+            quant_nf4=26 if "i4" in strategy else None,
+        )
+        self.w = {}  # fake weight
+        self.w["emb.weight"] = [0] * wrp.peek_info(model_path).num_vocab
+
+    def forward(self, tokens: List[int], state: Union[Any, None] = None):
+        return wrp.v5.run_one(self.model, tokens, state)

backend-python/utils/midi.py

@@ -52,6 +52,8 @@ class VocabConfig:
     bin_name_to_program_name: Dict[str, str]
     # Mapping from program number to instrument name.
     instrument_names: Dict[str, str]
+    # Manual override for velocity bins. Each element is the max velocity value for that bin by index.
+    velocity_bins_override: Optional[List[int]] = None
 
     def __post_init__(self):
         self.validate()
@@ -116,6 +118,12 @@ class VocabConfig:
             raise ValueError("velocity_bins must be at least 2")
         if len(self.bin_instrument_names) > 16:
             raise ValueError("bin_instruments must have at most 16 values")
+        if self.velocity_bins_override:
+            print("VocabConfig is using velocity_bins_override. Ignoring velocity_exp.")
+            if len(self.velocity_bins_override) != self.velocity_bins:
+                raise ValueError(
+                    "velocity_bins_override must have same length as velocity_bins"
+                )
         if (
             self.ch10_instrument_bin_name
             and self.ch10_instrument_bin_name not in self.bin_instrument_names
@@ -156,6 +164,11 @@ class VocabUtils:
 
     def velocity_to_bin(self, velocity: float) -> int:
         velocity = max(0, min(velocity, self.cfg.velocity_events - 1))
+        if self.cfg.velocity_bins_override:
+            for i, v in enumerate(self.cfg.velocity_bins_override):
+                if velocity <= v:
+                    return i
+            return 0
         binsize = self.cfg.velocity_events / (self.cfg.velocity_bins - 1)
         if self.cfg.velocity_exp == 1.0:
             return ceil(velocity / binsize)
@@ -176,6 +189,8 @@ class VocabUtils:
             )
 
     def bin_to_velocity(self, bin: int) -> int:
+        if self.cfg.velocity_bins_override:
+            return self.cfg.velocity_bins_override[bin]
         binsize = self.cfg.velocity_events / (self.cfg.velocity_bins - 1)
         if self.cfg.velocity_exp == 1.0:
             return max(0, ceil(bin * binsize - 1))
@@ -358,13 +373,32 @@ class AugmentConfig:
             )
 
 
+@dataclass
+class FilterConfig:
+    # Whether to filter out MIDI files with duplicate MD5 hashes.
+    deduplicate_md5: bool
+    # Minimum time delay between notes in a file before splitting into multiple documents.
+    piece_split_delay: float
+    # Minimum length of a piece in milliseconds.
+    min_piece_length: float
+
+    @classmethod
+    def from_json(cls, path: str):
+        with open(path, "r") as f:
+            config = json.load(f)
+        return cls(**config)
+
+
 def mix_volume(velocity: int, volume: int, expression: int) -> float:
     return velocity * (volume / 127.0) * (expression / 127.0)
 
 
 def convert_midi_to_str(
-    cfg: VocabConfig, mid: mido.MidiFile, augment: AugmentValues = None
-) -> str:
+    cfg: VocabConfig,
+    filter_cfg: FilterConfig,
+    mid: mido.MidiFile,
+    augment: AugmentValues = None,
+) -> List[str]:
     utils = VocabUtils(cfg)
     if augment is None:
         augment = AugmentValues.default()
@@ -390,7 +424,9 @@ def convert_midi_to_str(
     }  # {channel: {(note, program) -> True}}
     started_flag = False
 
+    output_list = []
     output = ["<start>"]
+    output_length_ms = 0.0
     token_data_buffer: List[
         Tuple[int, int, int, float]
     ] = []  # need to sort notes between wait tokens
@@ -432,16 +468,33 @@ def convert_midi_to_str(
         token_data_buffer = []
 
     def consume_note_program_data(prog: int, chan: int, note: int, vel: float):
-        nonlocal output, started_flag, delta_time_ms, cfg, utils, token_data_buffer
+        nonlocal output, output_length_ms, started_flag, delta_time_ms, cfg, utils, token_data_buffer
         is_token_valid = (
             utils.prog_data_to_token_data(prog, chan, note, vel) is not None
         )
         if not is_token_valid:
             return
+
+        if delta_time_ms > filter_cfg.piece_split_delay * 1000.0:
+            # check if any notes are still held
+            silent = True
+            for channel in channel_notes.keys():
+                if len(channel_notes[channel]) > 0:
+                    silent = False
+                    break
+            if silent:
+                flush_token_data_buffer()
+                output.append("<end>")
+                if output_length_ms > filter_cfg.min_piece_length * 1000.0:
+                    output_list.append(" ".join(output))
+                output = ["<start>"]
+                output_length_ms = 0.0
+                started_flag = False
         if started_flag:
             wait_tokens = utils.data_to_wait_tokens(delta_time_ms)
             if len(wait_tokens) > 0:
                 flush_token_data_buffer()
+                output_length_ms += delta_time_ms
                 output += wait_tokens
         delta_time_ms = 0.0
         token_data_buffer.append((prog, chan, note, vel * augment.velocity_mod_factor))
@@ -510,7 +563,9 @@ def convert_midi_to_str(
 
     flush_token_data_buffer()
     output.append("<end>")
-    return " ".join(output)
+    if output_length_ms > filter_cfg.min_piece_length * 1000.0:
+        output_list.append(" ".join(output))
+    return output_list
 
 
 def generate_program_change_messages(cfg: VocabConfig):
@@ -633,10 +688,10 @@ def token_to_midi_message(
                 if utils.cfg.decode_fix_repeated_notes:
                     if (channel, note) in state.active_notes:
                         del state.active_notes[(channel, note)]
-                    yield mido.Message(
-                        "note_off", note=note, time=ticks, channel=channel
-                    ), state
-                    ticks = 0
+                        yield mido.Message(
+                            "note_off", note=note, time=ticks, channel=channel
+                        ), state
+                        ticks = 0
                 state.active_notes[(channel, note)] = state.total_time
                 yield mido.Message(
                     "note_on", note=note, velocity=velocity, time=ticks, channel=channel

backend-python/utils/midi_filter_config.json

@@ -0,0 +1,5 @@
+{
+    "deduplicate_md5": true,
+    "piece_split_delay": 10.0,
+    "min_piece_length": 30.0
+}

backend-python/utils/rwkv.py

@@ -8,7 +8,6 @@ from typing import Dict, Iterable, List, Tuple, Union, Type
 from utils.log import quick_log
 from fastapi import HTTPException
 from pydantic import BaseModel, Field
-import numpy as np
 from routes import state_cache
 import global_var
 
@@ -68,6 +67,8 @@ class AbstractRWKV(ABC):
         pass
 
     def get_embedding(self, input: str, fast_mode: bool) -> Tuple[List[float], int]:
+        import numpy as np
+
         if fast_mode:
             embedding, token_len = self.__fast_embedding(
                 self.fix_tokens(self.pipeline.encode(input)), None
@@ -222,6 +223,8 @@ class AbstractRWKV(ABC):
     def generate(
         self, prompt: str, stop: Union[str, List[str], None] = None
     ) -> Iterable[Tuple[str, str, int, int]]:
+        import numpy as np
+
         quick_log(None, None, "Generation Prompt:\n" + prompt)
         cache = None
         delta_prompt = prompt
@@ -231,7 +234,7 @@ class AbstractRWKV(ABC):
             )
         except HTTPException:
             pass
-        if cache is None or cache["prompt"] == "":
+        if cache is None or cache["prompt"] == "" or cache["state"] is None:
             self.model_state = None
             self.model_tokens = []
         else:
@@ -510,12 +513,28 @@ def get_tokenizer(tokenizer_len: int):
 
 def RWKV(model: str, strategy: str, tokenizer: Union[str, None]) -> AbstractRWKV:
     rwkv_beta = global_var.get(global_var.Args).rwkv_beta
+    rwkv_cpp = getattr(global_var.get(global_var.Args), "rwkv.cpp")
+    webgpu = global_var.get(global_var.Args).webgpu
+
+    if "midi" in model.lower() or "abc" in model.lower():
+        os.environ["RWKV_RESCALE_LAYER"] = "999"
 
     # dynamic import to make RWKV_CUDA_ON work
     if rwkv_beta:
+        print("Using rwkv-beta")
         from rwkv_pip.beta.model import (
             RWKV as Model,
         )
+    elif rwkv_cpp:
+        print("Using rwkv.cpp, strategy is ignored")
+        from rwkv_pip.cpp.model import (
+            RWKV as Model,
+        )
+    elif webgpu:
+        print("Using webgpu")
+        from rwkv_pip.webgpu.model import (
+            RWKV as Model,
+        )
     else:
         from rwkv_pip.model import (
             RWKV as Model,

build/darwin/Readme_Install.txt

@@ -1,3 +1,8 @@
+Client Download URL:
+客户端下载地址:
+クライアントのダウンロードURL:
+https://github.com/josStorer/RWKV-Runner/releases/latest/download/RWKV-Runner_macos_universal.zip
+
 For Mac and Linux users, please manually install Python 3.10 (usually the latest systems come with it built-in). You can specify the Python interpreter to use in Settings. (which python3)
 对于Mac和Linux用户，请手动安装 Python3.10 (通常最新的系统已经内置了). 你可以在设置中指定使用的Python解释器. (which python3)
 MacおよびLinuxのユーザーの方は、Python3.10を手動でインストールしてください（通常、最新のシステムには既に組み込まれています）。 設定メニューで使用するPythonインタプリタを指定することができます。 (which python3)

build/linux/Readme_Install.txt

@@ -1,3 +1,8 @@
+Client Download URL:
+客户端下载地址:
+クライアントのダウンロードURL:
+https://github.com/josStorer/RWKV-Runner/releases/latest/download/RWKV-Runner_linux_x64
+
 For Mac and Linux users, please manually install Python 3.10 (usually the latest systems come with it built-in). You can specify the Python interpreter to use in Settings.
 对于Mac和Linux用户，请手动安装 Python3.10 (通常最新的系统已经内置了). 你可以在设置中指定使用的Python解释器.
 MacおよびLinuxのユーザーの方は、Python3.10を手動でインストールしてください（通常、最新のシステムには既に組み込まれています）。 設定メニューで使用するPythonインタプリタを指定することができます。

build/windows/Readme_Install.txt

@@ -1,3 +1,8 @@
+Client Download URL:
+客户端下载地址:
+クライアントのダウンロードURL:
+https://github.com/josStorer/RWKV-Runner/releases/latest/download/RWKV-Runner_windows_x64.exe
+
 Please execute this program in an empty directory. All related dependencies will be placed in this directory.
 请将本程序放在一个空目录内执行, 所有相关依赖均会放置于此目录.
 このプログラムを空のディレクトリで実行してください。関連するすべての依存関係は、このディレクトリに配置されます。

deploy-examples/ChatGPT-Next-Web/setup.sh

@@ -9,7 +9,7 @@ cd RWKV-Next-Web
 git clone https://github.com/josStorer/RWKV-Runner --depth=1
 python3 -m pip install torch torchvision torchaudio
 python3 -m pip install -r RWKV-Runner/backend-python/requirements.txt
-python3 ./RWKV-Runner/backend-python/main.py > log.txt &
+python3 ./RWKV-Runner/backend-python/main.py > log.txt & # this is only an example, you should use screen or other tools to run it in background
 
 if [ ! -d RWKV-Runner/models ]; then
     mkdir RWKV-Runner/models
@@ -22,6 +22,6 @@ yarn install
 yarn build
 export PROXY_URL=""
 export BASE_URL=http://127.0.0.1:8000
-yarn start &
+yarn start & # this is only an example, you should use screen or other tools to run it in background
 
 curl http://127.0.0.1:8000/switch-model -X POST -H "Content-Type: application/json" -d '{"model":"./RWKV-Runner/models/RWKV-4-World-0.1B-v1-20230520-ctx4096.pth","strategy":"cpu fp32"}'

deploy-examples/RWKV-Runner-WebUI/setup.bat

@@ -9,6 +9,7 @@ call npm ci
 call npm run build
 cd ..
 
+: optional: set ngrok_token=YOUR_NGROK_TOKEN
 start python ./backend-python/main.py --webui
 start "C:\Program Files (x86)\Microsoft\Edge\Application\msedge.exe" "http://127.0.0.1:8000"
 

deploy-examples/RWKV-Runner-WebUI/setup.sh

@@ -11,7 +11,8 @@ npm ci
 npm run build
 cd ..
 
-python3 ./backend-python/main.py --webui > log.txt &
+# optional: export ngrok_token=YOUR_NGROK_TOKEN
+python3 ./backend-python/main.py --webui > log.txt & # this is only an example, you should use screen or other tools to run it in background
 
 if [ ! -d models ]; then
     mkdir models

finetune/get_layer_and_embd.py

@@ -23,6 +23,7 @@ def file_cleaner(file):
     return cleaner
 
 
+expected_max_version = float(sys.argv[2]) if len(sys.argv) > 2 else 100
 model_file = open(sys.argv[1], "rb")
 cleaner = file_cleaner(model_file)
 cleaner_thread = threading.Thread(target=cleaner, daemon=True)
@@ -34,8 +35,23 @@ gc.collect()
 n_embd = w["emb.weight"].shape[1]
 n_layer = 0
 keys = list(w.keys())
+version = 4
 for x in keys:
     layer_id = int(x.split(".")[1]) if ("blocks." in x) else 0
     n_layer = max(n_layer, layer_id + 1)
 
-print(f"--n_layer {n_layer} --n_embd {n_embd}", end="")
+    if "ln_x" in x:
+        version = max(5, version)
+    if "gate.weight" in x:
+        version = max(5.1, version)
+    if int(version) == 5 and "att.time_decay" in x:
+        if len(w[x].shape) > 1:
+            if w[x].shape[1] > 1:
+                version = max(5.2, version)
+    if "time_maa" in x:
+        version = max(6, version)
+
+if version <= expected_max_version:
+    print(f"--n_layer {n_layer} --n_embd {n_embd}", end="")
+else:
+    raise Exception(f"RWKV{version} is not supported")

finetune/install-wsl-dep-and-train.sh

@@ -47,8 +47,12 @@ else
 fi
 
 echo "loading $loadModel"
-modelInfo=$(python3 ./finetune/get_layer_and_embd.py $loadModel)
+modelInfo=$(python3 ./finetune/get_layer_and_embd.py $loadModel 4)
 echo $modelInfo
-
-python3 ./finetune/lora/train.py $modelInfo $@ --proj_dir lora-models --data_type binidx --lora \
-  --lora_parts=att,ffn,time,ln --strategy deepspeed_stage_2 --accelerator gpu
+if [[ $modelInfo =~ "--n_layer" ]]; then
+  python3 ./finetune/lora/train.py $modelInfo $@ --proj_dir lora-models --data_type binidx --lora \
+    --lora_parts=att,ffn,time,ln --strategy deepspeed_stage_2 --accelerator gpu
+else
+  echo "modelInfo is invalid"
+  exit 1
+fi

frontend/package.json

@@ -21,10 +21,12 @@
     "i18next": "^22.4.15",
     "mobx": "^6.9.0",
     "mobx-react-lite": "^3.4.3",
+    "pdfjs-dist": "^4.0.189",
     "react": "^18.2.0",
     "react-beautiful-dnd": "^13.1.1",
     "react-chartjs-2": "^5.2.0",
     "react-dom": "^18.2.0",
+    "react-draggable": "^4.4.6",
     "react-i18next": "^12.2.2",
     "react-markdown": "^8.0.7",
     "react-router": "^6.11.1",
@@ -50,6 +52,7 @@
     "sass": "^1.62.1",
     "tailwindcss": "^3.3.2",
     "typescript": "^5.0.4",
-    "vite": "^4.3.6"
+    "vite": "^4.3.6",
+    "vite-plugin-top-level-await": "^1.3.1"
   }
 }

frontend/src/_locales/ja/main.json

@@ -82,7 +82,7 @@
   "Just like feeding sedatives to the model. Consider the results of the top n% probability mass, 0.1 considers the top 10%, with higher quality but more conservative, 1 considers all results, with lower quality but more diverse.": "モデルに鎮静剤を与えるようなもの。上位n％の確率質量の結果を考えてみてください。0.1は上位10％を考えており、質が高いが保守的で、1は全ての結果を考慮しており、質は低いが多様性があります。",
   "Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics.": "ポジティヴ値は、新しいトークンが今までのテキストに出現していたかどうかに基づいてこれらをペナルティとし、新しいトピックについて話す可能性を増加させます。",
   "Positive values penalize new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim.": "ポジティブ値は、新しいトークンが既存のテキストでどれだけ頻繁に使われているかに基づいてペナルティを与え、モデルが同じ行を完全に繰り返す可能性を減らします。",
-  "int8 uses less VRAM, but has slightly lower quality. fp16 has higher quality, and fp32 has the best quality.": "int8はVRAMの使用量が少ないですが、質が若干低いです。fp16は高品質、fp32は最高品質です。",
+  "int8 uses less VRAM, but has slightly lower quality. fp16 has higher quality.": "int8はVRAMの使用量が少ないですが、質が若干低いです。fp16は高品質。",
   "Number of the neural network layers loaded into VRAM, the more you load, the faster the speed, but it consumes more VRAM. (If your VRAM is not enough, it will fail to load)": "VRAMにロードされるニューラルネットワークの層の数。ロードする量が多いほど速度は速くなりますが、VRAMを多く消費します。(VRAMが不足している場合、ロードに失敗します)",
   "Whether to use CPU to calculate the last output layer of the neural network with FP32 precision to obtain better quality.": "ネットワークの最終出力層をFP32精度で計算するためにCPUを使用するかどうか。",
   "Downloads": "ダウンロード",
@@ -128,7 +128,7 @@
   "Chinese Kongfu": "中国武術",
   "Allow external access to the API (service must be restarted)": "APIへの外部アクセスを許可する (サービスを再起動する必要があります)",
   "Custom": "カスタム",
-  "CUDA (Beta, Faster)": "CUDA (ベータ、高速)",
+  "CUDA (Beta, Faster)": "CUDA (Beta, 高速)",
   "Reset All Configs": "すべての設定をリセット",
   "Cancel": "キャンセル",
   "Confirm": "確認",
@@ -229,7 +229,7 @@
   "Memory is not enough, try to increase the virtual memory (Swap of WSL) or use a smaller base model.": "メモリが不足しています、仮想メモリ (WSL Swap) を増やすか小さなベースモデルを使用してみてください。",
   "VRAM is not enough": "ビデオRAMが不足しています",
   "Training data is not enough, reduce context length or add more data for training": "トレーニングデータが不足しています、コンテキストの長さを減らすか、トレーニング用のデータをさらに追加してください",
-  "You are using WSL 1 for training, please upgrade to WSL 2. e.g. Run \"wsl --set-version Ubuntu-22.04 2\"": "トレーニングにWSL 1を使用しています、WSL 2にアップグレードしてください。例:\"wsl --set-version Ubuntu-22.04 2\"を実行する",
+  "Can not find an Nvidia GPU. Perhaps the gpu driver of windows is too old, or you are using WSL 1 for training, please upgrade to WSL 2. e.g. Run \"wsl --set-version Ubuntu-22.04 2\"": "Nvidia GPUが見つかりません。WindowsのGPUドライバが古すぎるか、トレーニングにWSL 1を使用している可能性があります。WSL 2にアップグレードしてください。例：\"wsl --set-version Ubuntu-22.04 2\"を実行してください",
   "Matched CUDA is not installed": "対応するCUDAがインストールされていません",
   "Failed to convert data": "データの変換に失敗しました",
   "Failed to merge model": "モデルのマージに失敗しました",
@@ -250,13 +250,13 @@
   "VRAM": "VRAM",
   "GPU Usage": "GPU使用率",
   "Use Custom Tokenizer": "カスタムトークナイザーを使用する",
-  "Tokenizer Path (e.g. backend-python/rwkv_pip/20B_tokenizer.json)": "トークナイザーパス (例: backend-python/rwkv_pip/20B_tokenizer.json)",
+  "Tokenizer Path (e.g. backend-python/rwkv_pip/20B_tokenizer.json or rwkv_vocab_v20230424.txt)": "トークナイザーパス (例: backend-python/rwkv_pip/20B_tokenizer.json または rwkv_vocab_v20230424.txt)",
   "User Name": "ユーザー名",
   "Assistant Name": "アシスタント名",
   "Insert default system prompt at the beginning": "最初にデフォルトのシステムプロンプトを挿入",
   "Format Content": "内容フォーマットの規格化",
   "Add An Attachment (Accepts pdf, txt)": "添付ファイルを追加 (pdf, txtを受け付けます)",
-  "Uploading Attachment": "添付ファイルアップロード中",
+  "Processing Attachment": "添付ファイルを処理中",
   "Remove Attachment": "添付ファイルを削除",
   "The content of file": "ファイル",
   "is as follows. When replying to me, consider the file content and respond accordingly:": "の内容は以下の通りです。私に返信する際は、ファイルの内容を考慮して適切に返信してください:",
@@ -268,5 +268,57 @@
   "Enable WebUI": "WebUIを有効化",
   "Server is working on deployment mode, please close the terminal window manually": "サーバーはデプロイモードで動作しています、ターミナルウィンドウを手動で閉じてください",
   "Server is working on deployment mode, please exit the program manually to stop the server": "サーバーはデプロイモードで動作しています、サーバーを停止するにはプログラムを手動で終了してください",
-  "You can increase the number of stored layers in Configs page to improve performance": "パフォーマンスを向上させるために、保存されるレイヤーの数を設定ページで増やすことができます"
+  "You can increase the number of stored layers in Configs page to improve performance": "パフォーマンスを向上させるために、保存されるレイヤーの数を設定ページで増やすことができます",
+  "Failed to load model, try to increase the virtual memory (Swap of WSL) or use a smaller base model.": "モデルの読み込みに失敗しました、仮想メモリ (WSL Swap) を増やすか小さなベースモデルを使用してみてください。",
+  "Save Conversation": "会話を保存",
+  "Use Hugging Face Mirror": "Hugging Faceミラーを使用",
+  "File is empty": "ファイルが空です",
+  "Open MIDI Input Audio Tracks": "MIDI入力オーディオトラックを開く",
+  "Track": "トラック",
+  "Play All": "すべて再生",
+  "Clear All": "すべてクリア",
+  "Scale View": "スケールビュー",
+  "Record": "録音",
+  "Play": "再生",
+  "New Track": "新規トラック",
+  "Select a track to preview the content": "トラックを選択して内容をプレビュー",
+  "Save to generation area": "生成エリアに保存",
+  "Piano": "ピアノ",
+  "Percussion": "パーカッション",
+  "Drum": "ドラム",
+  "Tuba": "チューバ",
+  "Marimba": "マリンバ",
+  "Bass": "ベース",
+  "Guitar": "ギター",
+  "Violin": "バイオリン",
+  "Trumpet": "トランペット",
+  "Sax": "サックス",
+  "Flute": "フルート",
+  "Lead": "リード",
+  "Pad": "パッド",
+  "MIDI Input": "MIDI入力",
+  "Select the MIDI input device to be used.": "使用するMIDI入力デバイスを選択します。",
+  "Start Time": "開始時間",
+  "Content Duration": "内容の長さ",
+  "Please select a MIDI device first": "まずMIDIデバイスを選択してください",
+  "Piano is the main instrument": "ピアノはメインの楽器です",
+  "Loss is too high, please check the training data, and ensure your gpu driver is up to date.": "Lossが大きすぎます、トレーニングデータを確認し、GPUドライバが最新であることを確認してください。",
+  "This version of RWKV is not supported yet.": "このバージョンのRWKVはまだサポートされていません。",
+  "Main": "メイン",
+  "Finetuned": "微調整",
+  "Global": "グローバル",
+  "Local": "ローカル",
+  "CN": "中国語",
+  "JP": "日本語",
+  "Music": "音楽",
+  "Other": "その他",
+  "Import MIDI": "MIDIをインポート",
+  "Current Instrument": "現在の楽器",
+  "Please convert model to GGML format first": "モデルをGGML形式に変換してください",
+  "Convert To GGML Format": "GGML形式に変換",
+  "CPU (rwkv.cpp, Faster)": "CPU (rwkv.cpp, 高速)",
+  "Play With External Player": "外部プレーヤーで再生",
+  "Core API URL": "コアAPI URL",
+  "Override core API URL(/chat/completions and /completions). If you don't know what this is, leave it blank.": "コアAPI URLを上書きします(/chat/completions と /completions)。何であるかわからない場合は空白のままにしてください。",
+  "Please change Strategy to CPU (rwkv.cpp) to use ggml format": "StrategyをCPU (rwkv.cpp)に変更して、ggml形式を使用してください"
 }

frontend/src/_locales/zh-hans/main.json

@@ -82,7 +82,7 @@
   "Just like feeding sedatives to the model. Consider the results of the top n% probability mass, 0.1 considers the top 10%, with higher quality but more conservative, 1 considers all results, with lower quality but more diverse.": "就像给模型喂镇静剂. 考虑前 n% 概率质量的结果, 0.1 考虑前 10%, 质量更高, 但更保守, 1 考虑所有质量结果, 质量降低, 但更多样",
   "Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics.": "存在惩罚. 正值根据新token在至今的文本中是否出现过, 来对其进行惩罚, 从而增加了模型涉及新话题的可能性",
   "Positive values penalize new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim.": "频率惩罚. 正值根据新token在至今的文本中出现的频率/次数, 来对其进行惩罚, 从而减少模型原封不动地重复相同句子的可能性",
-  "int8 uses less VRAM, but has slightly lower quality. fp16 has higher quality, and fp32 has the best quality.": "int8占用显存更低, 但质量略微下降. fp16质量更好, fp32质量最好",
+  "int8 uses less VRAM, but has slightly lower quality. fp16 has higher quality.": "int8占用显存更低, 但质量略微下降. fp16质量更好",
   "Number of the neural network layers loaded into VRAM, the more you load, the faster the speed, but it consumes more VRAM. (If your VRAM is not enough, it will fail to load)": "载入显存的神经网络层数, 载入越多, 速度越快, 但显存消耗越大 (如果你的显存不够, 会载入失败)",
   "Whether to use CPU to calculate the last output layer of the neural network with FP32 precision to obtain better quality.": "是否使用cpu以fp32精度计算神经网络的最后一层输出层, 以获得更好的质量",
   "Downloads": "下载",
@@ -229,7 +229,7 @@
   "Memory is not enough, try to increase the virtual memory (Swap of WSL) or use a smaller base model.": "内存不足，尝试增加虚拟内存(WSL Swap)，或使用一个更小规模的基底模型",
   "VRAM is not enough": "显存不足",
   "Training data is not enough, reduce context length or add more data for training": "训练数据不足，请减小上下文长度或增加训练数据",
-  "You are using WSL 1 for training, please upgrade to WSL 2. e.g. Run \"wsl --set-version Ubuntu-22.04 2\"": "你正在使用WSL 1进行训练，请升级到WSL 2。例如，运行\"wsl --set-version Ubuntu-22.04 2\"",
+  "Can not find an Nvidia GPU. Perhaps the gpu driver of windows is too old, or you are using WSL 1 for training, please upgrade to WSL 2. e.g. Run \"wsl --set-version Ubuntu-22.04 2\"": "没有找到Nvidia显卡。可能是因为你的windows显卡驱动太旧，或者你正在使用WSL 1进行训练，请升级到WSL 2。例如，执行\"wsl --set-version Ubuntu-22.04 2\"",
   "Matched CUDA is not installed": "未安装匹配的CUDA",
   "Failed to convert data": "数据转换失败",
   "Failed to merge model": "合并模型失败",
@@ -250,13 +250,13 @@
   "VRAM": "显存",
   "GPU Usage": "GPU占用",
   "Use Custom Tokenizer": "使用自定义Tokenizer",
-  "Tokenizer Path (e.g. backend-python/rwkv_pip/20B_tokenizer.json)": "Tokenizer路径 (例如: backend-python/rwkv_pip/20B_tokenizer.json)",
+  "Tokenizer Path (e.g. backend-python/rwkv_pip/20B_tokenizer.json or rwkv_vocab_v20230424.txt)": "Tokenizer路径 (例如: backend-python/rwkv_pip/20B_tokenizer.json 或 rwkv_vocab_v20230424.txt)",
   "User Name": "用户名称",
   "Assistant Name": "AI名称",
   "Insert default system prompt at the beginning": "在开头自动插入默认系统提示",
   "Format Content": "规范格式",
   "Add An Attachment (Accepts pdf, txt)": "添加一个附件 (支持pdf, txt)",
-  "Uploading Attachment": "正在上传附件",
+  "Processing Attachment": "正在处理附件",
   "Remove Attachment": "移除附件",
   "The content of file": "文件",
   "is as follows. When replying to me, consider the file content and respond accordingly:": "内容如下。回复时考虑文件内容并做出相应回复:",
@@ -268,5 +268,57 @@
   "Enable WebUI": "启用WebUI",
   "Server is working on deployment mode, please close the terminal window manually": "服务器正在部署模式下运行，请手动关闭终端窗口",
   "Server is working on deployment mode, please exit the program manually to stop the server": "服务器正在部署模式下运行，请手动退出程序以停止服务器",
-  "You can increase the number of stored layers in Configs page to improve performance": "你可以在配置页面增加载入显存层数以提升性能"
+  "You can increase the number of stored layers in Configs page to improve performance": "你可以在配置页面增加载入显存层数以提升性能",
+  "Failed to load model, try to increase the virtual memory (Swap of WSL) or use a smaller base model.": "模型载入失败，尝试增加虚拟内存(WSL Swap)，或使用一个更小规模的基底模型",
+  "Save Conversation": "保存对话",
+  "Use Hugging Face Mirror": "使用Hugging Face镜像源",
+  "File is empty": "文件为空",
+  "Open MIDI Input Audio Tracks": "打开MIDI输入音轨",
+  "Track": "音轨",
+  "Play All": "播放全部",
+  "Clear All": "清空",
+  "Scale View": "缩放视图",
+  "Record": "录制",
+  "Play": "播放",
+  "New Track": "新建音轨",
+  "Select a track to preview the content": "选择一个音轨以预览内容",
+  "Save to generation area": "保存到生成区",
+  "Piano": "钢琴",
+  "Percussion": "打击乐",
+  "Drum": "鼓",
+  "Tuba": "大号",
+  "Marimba": "马林巴",
+  "Bass": "贝斯",
+  "Guitar": "吉他",
+  "Violin": "小提琴",
+  "Trumpet": "小号",
+  "Sax": "萨克斯",
+  "Flute": "长笛",
+  "Lead": "主音",
+  "Pad": "和音",
+  "MIDI Input": "MIDI输入",
+  "Select the MIDI input device to be used.": "选择要使用的MIDI输入设备",
+  "Start Time": "开始时间",
+  "Content Duration": "内容时长",
+  "Please select a MIDI device first": "请先选择一个MIDI设备",
+  "Piano is the main instrument": "钢琴为主",
+  "Loss is too high, please check the training data, and ensure your gpu driver is up to date.": "Loss过高，请检查训练数据，并确保你的显卡驱动是最新的",
+  "This version of RWKV is not supported yet.": "暂不支持此版本的RWKV",
+  "Main": "主干",
+  "Finetuned": "微调",
+  "Global": "全球",
+  "Local": "本地",
+  "CN": "中文",
+  "JP": "日文",
+  "Music": "音乐",
+  "Other": "其他",
+  "Import MIDI": "导入MIDI",
+  "Current Instrument": "当前乐器",
+  "Please convert model to GGML format first": "请先将模型转换为GGML格式",
+  "Convert To GGML Format": "转换为GGML格式",
+  "CPU (rwkv.cpp, Faster)": "CPU (rwkv.cpp, 更快)",
+  "Play With External Player": "使用外部播放器播放",
+  "Core API URL": "核心 API URL",
+  "Override core API URL(/chat/completions and /completions). If you don't know what this is, leave it blank.": "覆盖核心的 API URL (/chat/completions 和 /completions)。如果你不知道这是什么，请留空",
+  "Please change Strategy to CPU (rwkv.cpp) to use ggml format": "请将Strategy改为CPU (rwkv.cpp)以使用ggml格式"
 }

frontend/src/components/ConfigSelector.tsx

@@ -1,10 +1,10 @@
 import { FC } from 'react';
 import { observer } from 'mobx-react-lite';
-import { Dropdown, Option } from '@fluentui/react-components';
+import { Dropdown, Option, PresenceBadge } from '@fluentui/react-components';
 import commonStore from '../stores/commonStore';
 
 export const ConfigSelector: FC<{ size?: 'small' | 'medium' | 'large' }> = observer(({ size }) => {
-  return <Dropdown size={size} style={{ minWidth: 0 }} listbox={{ style: { minWidth: 0 } }}
+  return <Dropdown size={size} style={{ minWidth: 0 }} listbox={{ style: { minWidth: 'fit-content' } }}
     value={commonStore.getCurrentModelConfig().name}
     selectedOptions={[commonStore.currentModelConfigIndex.toString()]}
     onOptionSelect={(_, data) => {
@@ -12,7 +12,13 @@ export const ConfigSelector: FC<{ size?: 'small' | 'medium' | 'large' }> = obser
         commonStore.setCurrentConfigIndex(Number(data.optionValue));
     }}>
     {commonStore.modelConfigs.map((config, index) =>
-      <Option key={index} value={index.toString()}>{config.name}</Option>
+      <Option key={index} value={index.toString()} text={config.name}>
+        <div className="flex justify-between grow">
+          {config.name}
+          {commonStore.modelSourceList.find(item => item.name === config.modelParameters.modelName)?.isComplete
+            && <PresenceBadge status="available" />}
+        </div>
+      </Option>
     )}
   </Dropdown>;
 });

frontend/src/components/LazyImportComponent.tsx

@@ -1,5 +1,6 @@
 import { FC, LazyExoticComponent, ReactNode, Suspense } from 'react';
 import { useTranslation } from 'react-i18next';
+import { Spinner } from '@fluentui/react-components';
 
 interface LazyImportComponentProps {
   lazyChildren: LazyExoticComponent<FC<any>>;
@@ -11,7 +12,10 @@ export const LazyImportComponent: FC<LazyImportComponentProps> = (props) => {
   const { t } = useTranslation();
 
   return (
-    <Suspense fallback={<div>{t('Loading...')}</div>}>
+    <Suspense fallback={
+      <div className="flex justify-center items-center h-full w-full">
+        <Spinner size="huge" label={t('Loading...')} />
+      </div>}>
       <props.lazyChildren {...props.lazyProps}>
         {props.children}
       </props.lazyChildren>

frontend/src/components/RunButton.tsx

@@ -11,12 +11,14 @@ import { Button } from '@fluentui/react-components';
 import { observer } from 'mobx-react-lite';
 import { exit, getStatus, readRoot, switchModel, updateConfig } from '../apis';
 import { toast } from 'react-toastify';
-import { checkDependencies, getStrategy, toastWithButton } from '../utils';
+import { checkDependencies, getHfDownloadUrl, getStrategy, toastWithButton } from '../utils';
 import { useTranslation } from 'react-i18next';
 import { ToolTipButton } from './ToolTipButton';
 import { Play16Regular, Stop16Regular } from '@fluentui/react-icons';
 import { useNavigate } from 'react-router';
 import { WindowShow } from '../../wailsjs/runtime';
+import { convertToGGML, convertToSt } from '../utils/convert-model';
+import { Precision } from '../types/configs';
 
 const mainButtonText = {
   [ModelStatus.Offline]: 'Run',
@@ -46,6 +48,8 @@ export const RunButton: FC<{ onClickRun?: MouseEventHandler, iconMode?: boolean
 
       const modelConfig = commonStore.getCurrentModelConfig();
       const webgpu = modelConfig.modelParameters.device === 'WebGPU';
+      const webgpuPython = modelConfig.modelParameters.device === 'WebGPU (Python)';
+      const cpp = modelConfig.modelParameters.device === 'CPU (rwkv.cpp)';
       let modelName = '';
       let modelPath = '';
       if (modelConfig && modelConfig.modelParameters) {
@@ -57,20 +61,47 @@ export const RunButton: FC<{ onClickRun?: MouseEventHandler, iconMode?: boolean
         return;
       }
 
-      if (webgpu) {
+      const currentModelSource = commonStore.modelSourceList.find(item => item.name === modelName);
+
+      const showDownloadPrompt = (promptInfo: string, downloadName: string) => {
+        toastWithButton(promptInfo, t('Download'), () => {
+          const downloadUrl = currentModelSource?.downloadUrl;
+          if (downloadUrl) {
+            toastWithButton(`${t('Downloading')} ${downloadName}`, t('Check'), () => {
+                navigate({ pathname: '/downloads' });
+              },
+              { autoClose: 3000 });
+            AddToDownloadList(modelPath, getHfDownloadUrl(downloadUrl));
+          } else {
+            toast(t('Can not find download url'), { type: 'error' });
+          }
+        });
+      };
+
+      if (webgpu || webgpuPython) {
         if (!['.st', '.safetensors'].some(ext => modelPath.endsWith(ext))) {
           const stModelPath = modelPath.replace(/\.pth$/, '.st');
           if (await FileExists(stModelPath)) {
             modelPath = stModelPath;
+          } else if (!await FileExists(modelPath)) {
+            showDownloadPrompt(t('Model file not found'), modelName);
+            commonStore.setStatus({ status: ModelStatus.Offline });
+            return;
+          } else if (!currentModelSource?.isComplete) {
+            showDownloadPrompt(t('Model file download is not complete'), modelName);
+            commonStore.setStatus({ status: ModelStatus.Offline });
+            return;
           } else {
-            toast(t('Please convert model to safe tensors format first'), { type: 'error' });
+            toastWithButton(t('Please convert model to safe tensors format first'), t('Convert'), () => {
+              convertToSt(modelConfig, navigate);
+            });
             commonStore.setStatus({ status: ModelStatus.Offline });
             return;
           }
         }
       }
 
-      if (!webgpu) {
+      if (!webgpu && !webgpuPython) {
         if (['.st', '.safetensors'].some(ext => modelPath.endsWith(ext))) {
           toast(t('Please change Strategy to WebGPU to use safetensors format'), { type: 'error' });
           commonStore.setStatus({ status: ModelStatus.Offline });
@@ -84,28 +115,45 @@ export const RunButton: FC<{ onClickRun?: MouseEventHandler, iconMode?: boolean
           return;
       }
 
-      const currentModelSource = commonStore.modelSourceList.find(item => item.name === modelName);
-
-      const showDownloadPrompt = (promptInfo: string, downloadName: string) => {
-        toastWithButton(promptInfo, t('Download'), () => {
-          const downloadUrl = currentModelSource?.downloadUrl;
-          if (downloadUrl) {
-            toastWithButton(`${t('Downloading')} ${downloadName}`, t('Check'), () => {
-                navigate({ pathname: '/downloads' });
-              },
-              { autoClose: 3000 });
-            AddToDownloadList(modelPath, downloadUrl);
+      if (cpp) {
+        if (!['.bin'].some(ext => modelPath.endsWith(ext))) {
+          const precision: Precision = modelConfig.modelParameters.precision === 'Q5_1' ? 'Q5_1' : 'fp16';
+          const ggmlModelPath = modelPath.replace(/\.pth$/, `-${precision}.bin`);
+          if (await FileExists(ggmlModelPath)) {
+            modelPath = ggmlModelPath;
+          } else if (!await FileExists(modelPath)) {
+            showDownloadPrompt(t('Model file not found'), modelName);
+            commonStore.setStatus({ status: ModelStatus.Offline });
+            return;
+          } else if (!currentModelSource?.isComplete) {
+            showDownloadPrompt(t('Model file download is not complete'), modelName);
+            commonStore.setStatus({ status: ModelStatus.Offline });
+            return;
           } else {
-            toast(t('Can not find download url'), { type: 'error' });
+            toastWithButton(t('Please convert model to GGML format first'), t('Convert'), () => {
+              convertToGGML(modelConfig, navigate);
+            });
+            commonStore.setStatus({ status: ModelStatus.Offline });
+            return;
           }
-        });
-      };
+        }
+      }
+
+      if (!cpp) {
+        if (['.bin'].some(ext => modelPath.endsWith(ext))) {
+          toast(t('Please change Strategy to CPU (rwkv.cpp) to use ggml format'), { type: 'error' });
+          commonStore.setStatus({ status: ModelStatus.Offline });
+          return;
+        }
+      }
 
       if (!await FileExists(modelPath)) {
         showDownloadPrompt(t('Model file not found'), modelName);
         commonStore.setStatus({ status: ModelStatus.Offline });
         return;
-      } else if (!currentModelSource?.isComplete) {
+      } else // If the user selects the .pth model with WebGPU mode, modelPath will be set to the .st model.
+        // However, if the .pth model is deleted, modelPath will exist and isComplete will be false.
+      if (!currentModelSource?.isComplete && modelPath.endsWith('.pth')) {
         showDownloadPrompt(t('Model file download is not complete'), modelName);
         commonStore.setStatus({ status: ModelStatus.Offline });
         return;
@@ -129,7 +177,7 @@ export const RunButton: FC<{ onClickRun?: MouseEventHandler, iconMode?: boolean
       const isUsingCudaBeta = modelConfig.modelParameters.device === 'CUDA-Beta';
 
       startServer(commonStore.settings.customPythonPath, port, commonStore.settings.host !== '127.0.0.1' ? '0.0.0.0' : '127.0.0.1',
-        !!modelConfig.enableWebUI, isUsingCudaBeta
+        !!modelConfig.enableWebUI, isUsingCudaBeta, cpp, webgpuPython
       ).catch((e) => {
         const errMsg = e.message || e;
         if (errMsg.includes('path contains space'))
@@ -138,6 +186,8 @@ export const RunButton: FC<{ onClickRun?: MouseEventHandler, iconMode?: boolean
           toast(t('Error') + ' - ' + errMsg, { type: 'error' });
       });
       setTimeout(WindowShow, 1000);
+      setTimeout(WindowShow, 2000);
+      setTimeout(WindowShow, 3000);
 
       let timeoutCount = 6;
       let loading = false;
@@ -154,7 +204,7 @@ export const RunButton: FC<{ onClickRun?: MouseEventHandler, iconMode?: boolean
               });
             }
             commonStore.setStatus({ status: ModelStatus.Loading });
-            toast(t('Loading Model'), { type: 'info' });
+            const loadingId = toast(t('Loading Model'), { type: 'info', autoClose: false });
             if (!webgpu) {
               updateConfig({
                 max_tokens: modelConfig.apiParameters.maxResponseToken,
@@ -167,8 +217,9 @@ export const RunButton: FC<{ onClickRun?: MouseEventHandler, iconMode?: boolean
 
             const strategy = getStrategy(modelConfig);
             let customCudaFile = '';
-            if ((modelConfig.modelParameters.device.includes('CUDA') || modelConfig.modelParameters.device === 'Custom')
-              && modelConfig.modelParameters.useCustomCuda && !strategy.includes('fp32')) {
+            if ((modelConfig.modelParameters.device.startsWith('CUDA') || modelConfig.modelParameters.device === 'Custom')
+              && modelConfig.modelParameters.useCustomCuda
+              && !strategy.split('->').some(s => ['cuda', 'fp32'].every(v => s.includes(v)))) {
               if (commonStore.platform === 'windows') {
                 // this part is currently unused because there's no longer a need to use different kernels for different GPUs, but it might still be needed in the future
                 //
@@ -214,10 +265,10 @@ export const RunButton: FC<{ onClickRun?: MouseEventHandler, iconMode?: boolean
                   navigate({ pathname: '/' + buttonName.toLowerCase() });
                 };
 
-                if ((modelConfig.modelParameters.device === 'CUDA' || modelConfig.modelParameters.device === 'CUDA-Beta') &&
+                if (modelConfig.modelParameters.device.startsWith('CUDA') &&
                   modelConfig.modelParameters.storedLayers < modelConfig.modelParameters.maxStoredLayers &&
                   commonStore.monitorData && commonStore.monitorData.totalVram !== 0 &&
-                  (commonStore.monitorData.usedVram / commonStore.monitorData.totalVram) < 0.85)
+                  (commonStore.monitorData.usedVram / commonStore.monitorData.totalVram) < 0.9)
                   toast(t('You can increase the number of stored layers in Configs page to improve performance'), { type: 'info' });
                 toastWithButton(t('Startup Completed'), t(buttonName), buttonFn, { type: 'success', autoClose: 3000 });
               } else if (r.status === 304) {
@@ -240,6 +291,8 @@ export const RunButton: FC<{ onClickRun?: MouseEventHandler, iconMode?: boolean
             }).catch((e) => {
               commonStore.setStatus({ status: ModelStatus.Offline });
               toast(t('Failed to switch model') + ' - ' + (e.message || e), { type: 'error' });
+            }).finally(() => {
+              toast.dismiss(loadingId);
             });
           }
         }).catch(() => {

frontend/src/components/WorkHeader.tsx

@@ -6,6 +6,7 @@ import { ConfigSelector } from './ConfigSelector';
 import { RunButton } from './RunButton';
 import { PresenceBadgeStatus } from '@fluentui/react-badge';
 import { useTranslation } from 'react-i18next';
+import { useMediaQuery } from 'usehooks-ts';
 
 const statusText = {
   [ModelStatus.Offline]: 'Offline',
@@ -23,6 +24,7 @@ const badgeStatus: { [modelStatus: number]: PresenceBadgeStatus } = {
 
 export const WorkHeader: FC = observer(() => {
   const { t } = useTranslation();
+  const mq = useMediaQuery('(min-width: 640px)');
   const port = commonStore.getCurrentModelConfig().apiParameters.apiPort;
 
   return commonStore.platform === 'web' ?
@@ -33,6 +35,10 @@ export const WorkHeader: FC = observer(() => {
           <PresenceBadge status={badgeStatus[commonStore.status.status]} />
           <Text size={100}>{t('Model Status') + ': ' + t(statusText[commonStore.status.status])}</Text>
         </div>
+        {commonStore.lastModelName && mq &&
+          <Text size={100}>
+            {commonStore.lastModelName}
+          </Text>}
         <div className="flex items-center gap-2">
           <ConfigSelector size="small" />
           <RunButton iconMode />

frontend/src/pages/AudiotrackManager/AudiotrackButton.tsx

@@ -0,0 +1,40 @@
+import React, { FC, lazy } from 'react';
+import { useTranslation } from 'react-i18next';
+import { Button, Dialog, DialogBody, DialogContent, DialogSurface, DialogTrigger } from '@fluentui/react-components';
+import { CustomToastContainer } from '../../components/CustomToastContainer';
+import { LazyImportComponent } from '../../components/LazyImportComponent';
+import { flushMidiRecordingContent } from '../../utils';
+import commonStore from '../../stores/commonStore';
+
+const AudiotrackEditor = lazy(() => import('./AudiotrackEditor'));
+
+export const AudiotrackButton: FC<{
+  size?: 'small' | 'medium' | 'large',
+  shape?: 'rounded' | 'circular' | 'square';
+  appearance?: 'secondary' | 'primary' | 'outline' | 'subtle' | 'transparent';
+  setPrompt: (prompt: string) => void;
+}> = ({ size, shape, appearance, setPrompt }) => {
+  const { t } = useTranslation();
+
+  return <Dialog onOpenChange={(e, data) => {
+    if (!data.open) {
+      flushMidiRecordingContent();
+      commonStore.setRecordingTrackId('');
+      commonStore.setPlayingTrackId('');
+    }
+  }}>
+    <DialogTrigger disableButtonEnhancement>
+      <Button size={size} shape={shape} appearance={appearance}>
+        {t('Open MIDI Input Audio Tracks')}
+      </Button>
+    </DialogTrigger>
+    <DialogSurface style={{ paddingTop: 0, maxWidth: '90vw', width: 'fit-content' }}>
+      <DialogBody>
+        <DialogContent className="overflow-hidden">
+          <CustomToastContainer />
+          <LazyImportComponent lazyChildren={AudiotrackEditor} lazyProps={{ setPrompt }} />
+        </DialogContent>
+      </DialogBody>
+    </DialogSurface>
+  </Dialog>;
+};

frontend/src/pages/AudiotrackManager/AudiotrackEditor.tsx

@@ -0,0 +1,601 @@
+import React, { FC, useEffect, useRef, useState } from 'react';
+import { observer } from 'mobx-react-lite';
+import { useTranslation } from 'react-i18next';
+import Draggable from 'react-draggable';
+import { ToolTipButton } from '../../components/ToolTipButton';
+import { v4 as uuid } from 'uuid';
+import {
+  Add16Regular,
+  ArrowAutofitWidth20Regular,
+  ArrowUpload16Regular,
+  Delete16Regular,
+  MusicNote220Regular,
+  Pause16Regular,
+  Play16Filled,
+  Play16Regular,
+  Record16Regular,
+  Stop16Filled
+} from '@fluentui/react-icons';
+import { Button, Card, DialogTrigger, Slider, Text, Tooltip } from '@fluentui/react-components';
+import { useWindowSize } from 'usehooks-ts';
+import commonStore, { ModelStatus } from '../../stores/commonStore';
+import classnames from 'classnames';
+import {
+  InstrumentType,
+  InstrumentTypeNameMap,
+  InstrumentTypeTokenMap,
+  MidiMessage,
+  tracksMinimalTotalTime
+} from '../../types/composition';
+import { toast } from 'react-toastify';
+import {
+  absPathAsset,
+  flushMidiRecordingContent,
+  getMidiRawContentMainInstrument,
+  getMidiRawContentTime,
+  getServerRoot,
+  refreshTracksTotalTime
+} from '../../utils';
+import { OpenOpenFileDialog, PlayNote } from '../../../wailsjs/go/backend_golang/App';
+
+const snapValue = 25;
+const minimalMoveTime = 8; // 1000/125=8ms wait_events=125
+const scaleMin = 0.05;
+const scaleMax = 3;
+const baseMoveTime = Math.round(minimalMoveTime / scaleMin);
+
+const velocityEvents = 128;
+const velocityBins = 12;
+const velocityExp = 0.5;
+
+const minimalTrackWidth = 80;
+const trackInitOffsetPx = 10;
+const pixelFix = 0.5;
+const topToArrowIcon = 19;
+const arrowIconToTracks = 23;
+
+const velocityToBin = (velocity: number) => {
+  velocity = Math.max(0, Math.min(velocity, velocityEvents - 1));
+  const binsize = velocityEvents / (velocityBins - 1);
+  return Math.ceil((velocityEvents * ((Math.pow(velocityExp, (velocity / velocityEvents)) - 1.0) / (velocityExp - 1.0))) / binsize);
+};
+
+const binToVelocity = (bin: number) => {
+  const binsize = velocityEvents / (velocityBins - 1);
+  return Math.max(0, Math.ceil(velocityEvents * (Math.log(((velocityExp - 1) * binsize * bin) / velocityEvents + 1) / Math.log(velocityExp)) - 1));
+};
+
+const tokenToMidiMessage = (token: string): MidiMessage | null => {
+  if (token.startsWith('<')) return null;
+  if (token.startsWith('t') && !token.startsWith('t:')) {
+    return {
+      messageType: 'ElapsedTime',
+      value: parseInt(token.substring(1)) * minimalMoveTime,
+      channel: 0,
+      note: 0,
+      velocity: 0,
+      control: 0,
+      instrument: 0
+    };
+  }
+  const instrument: InstrumentType = InstrumentTypeTokenMap.findIndex(t => token.startsWith(t + ':'));
+  if (instrument >= 0) {
+    const parts = token.split(':');
+    if (parts.length !== 3) return null;
+    const note = parseInt(parts[1], 16);
+    const velocity = parseInt(parts[2], 16);
+    if (velocity < 0 || velocity > 127) return null;
+    if (velocity === 0) return {
+      messageType: 'NoteOff',
+      note: note,
+      instrument: instrument,
+      channel: 0,
+      velocity: 0,
+      control: 0,
+      value: 0
+    };
+    return {
+      messageType: 'NoteOn',
+      note: note,
+      velocity: binToVelocity(velocity),
+      instrument: instrument,
+      channel: 0,
+      control: 0,
+      value: 0
+    } as MidiMessage;
+  }
+  return null;
+};
+
+const midiMessageToToken = (msg: MidiMessage) => {
+  if (msg.messageType === 'NoteOn' || msg.messageType === 'NoteOff') {
+    const instrument = InstrumentTypeTokenMap[msg.instrument];
+    const note = msg.note.toString(16);
+    const velocity = velocityToBin(msg.velocity).toString(16);
+    return `${instrument}:${note}:${velocity} `;
+  } else if (msg.messageType === 'ElapsedTime') {
+    let time = Math.round(msg.value / minimalMoveTime);
+    const num = Math.floor(time / 125); // wait_events=125
+    time -= num * 125;
+    let ret = '';
+    for (let i = 0; i < num; i++) {
+      ret += 't125 ';
+    }
+    if (time > 0)
+      ret += `t${time} `;
+    return ret;
+  } else
+    return '';
+};
+
+let dropRecordingTime = false;
+
+export const midiMessageHandler = async (data: MidiMessage) => {
+  if (data.messageType === 'ControlChange') {
+    commonStore.setInstrumentType(Math.round(data.value / 127 * (InstrumentTypeNameMap.length - 1)));
+    return;
+  }
+  if (commonStore.recordingTrackId) {
+    if (dropRecordingTime && data.messageType === 'ElapsedTime') {
+      dropRecordingTime = false;
+      return;
+    }
+    data = {
+      ...data,
+      instrument: commonStore.instrumentType
+    };
+    commonStore.setRecordingRawContent([...commonStore.recordingRawContent, data]);
+    commonStore.setRecordingContent(commonStore.recordingContent + midiMessageToToken(data));
+
+    //TODO data.channel = data.instrument;
+    PlayNote(data);
+  }
+};
+
+type TrackProps = {
+  id: string;
+  right: number;
+  scale: number;
+  isSelected: boolean;
+  onSelect: (id: string) => void;
+};
+
+const Track: React.FC<TrackProps> = observer(({
+  id,
+  right,
+  scale,
+  isSelected,
+  onSelect
+}) => {
+  const { t } = useTranslation();
+  const trackIndex = commonStore.tracks.findIndex(t => t.id === id)!;
+  const track = commonStore.tracks[trackIndex];
+  const trackClass = isSelected ? 'bg-blue-600' : (commonStore.settings.darkMode ? 'bg-blue-900' : 'bg-gray-700');
+  const controlX = useRef(0);
+
+  let trackName = t('Track') + ' ' + id;
+  if (track.mainInstrument)
+    trackName = t('Track') + ' - ' + t('Piano is the main instrument')!.replace(t('Piano')!, t(track.mainInstrument)) + (track.content && (' - ' + track.content));
+  else if (track.content)
+    trackName = t('Track') + ' - ' + track.content;
+
+  return (
+    <Draggable
+      axis="x"
+      bounds={{ left: 0, right }}
+      grid={[snapValue, snapValue]}
+      position={{
+        x: (track.offsetTime - commonStore.trackCurrentTime) / (baseMoveTime * scale) * snapValue,
+        y: 0
+      }}
+      onStart={(e, data) => {
+        controlX.current = data.lastX;
+      }}
+      onStop={(e, data) => {
+        const delta = data.lastX - controlX.current;
+        let offsetTime = Math.round(Math.round(delta / snapValue * baseMoveTime * scale) / minimalMoveTime) * minimalMoveTime;
+        offsetTime = Math.min(Math.max(
+          offsetTime,
+          -track.offsetTime), commonStore.trackTotalTime - track.offsetTime);
+
+        const tracks = commonStore.tracks.slice();
+        tracks[trackIndex].offsetTime += offsetTime;
+        commonStore.setTracks(tracks);
+        refreshTracksTotalTime();
+      }}
+    >
+      <div
+        className={`p-1 cursor-move rounded whitespace-nowrap overflow-hidden ${trackClass}`}
+        style={{
+          width: `${Math.max(minimalTrackWidth,
+            track.contentTime / (baseMoveTime * scale) * snapValue
+          )}px`
+        }}
+        onClick={() => onSelect(id)}
+      >
+        <span className="text-white">{trackName}</span>
+      </div>
+    </Draggable>
+  );
+});
+
+const AudiotrackEditor: FC<{ setPrompt: (prompt: string) => void }> = observer(({ setPrompt }) => {
+  const { t } = useTranslation();
+
+  const viewControlsContainerRef = useRef<HTMLDivElement>(null);
+  const currentTimeControlRef = useRef<HTMLDivElement>(null);
+  const playStartTimeControlRef = useRef<HTMLDivElement>(null);
+  const tracksEndLineRef = useRef<HTMLDivElement>(null);
+  const tracksRef = useRef<HTMLDivElement>(null);
+  const toolbarRef = useRef<HTMLDivElement>(null);
+  const toolbarButtonRef = useRef<HTMLDivElement>(null);
+  const toolbarSliderRef = useRef<HTMLInputElement>(null);
+  const contentPreviewRef = useRef<HTMLDivElement>(null);
+
+  const [refreshRef, setRefreshRef] = useState(false);
+
+  const windowSize = useWindowSize();
+  const scale = (scaleMin + scaleMax) - commonStore.trackScale;
+
+  const [selectedTrackId, setSelectedTrackId] = useState<string>('');
+  const playStartTimeControlX = useRef(0);
+  const selectedTrack = selectedTrackId ? commonStore.tracks.find(t => t.id === selectedTrackId) : undefined;
+
+  useEffect(() => {
+    if (toolbarSliderRef.current && toolbarSliderRef.current.parentElement)
+      toolbarSliderRef.current.parentElement.style.removeProperty('--fui-Slider--steps-percent');
+  }, []);
+
+  const scrollContentToBottom = () => {
+    if (contentPreviewRef.current)
+      contentPreviewRef.current.scrollTop = contentPreviewRef.current.scrollHeight;
+  };
+
+  useEffect(() => {
+    scrollContentToBottom();
+  }, [commonStore.recordingContent]);
+
+  useEffect(() => {
+    setRefreshRef(!refreshRef);
+  }, [windowSize, commonStore.tracks]);
+
+  const viewControlsContainerWidth = (toolbarRef.current && toolbarButtonRef.current && toolbarSliderRef.current) ?
+    toolbarRef.current.clientWidth - toolbarButtonRef.current.clientWidth - toolbarSliderRef.current.clientWidth - 16 // 16 = ml-2 mr-2
+    : 0;
+  const tracksWidth = viewControlsContainerWidth;
+  const timeOfTracksWidth = Math.floor(tracksWidth / snapValue) // number of moves
+    * baseMoveTime * scale;
+  const currentTimeControlWidth = (timeOfTracksWidth < commonStore.trackTotalTime)
+    ? timeOfTracksWidth / commonStore.trackTotalTime * viewControlsContainerWidth
+    : 0;
+  const playStartTimeControlPosition = (commonStore.trackPlayStartTime - commonStore.trackCurrentTime) / (baseMoveTime * scale) * snapValue;
+  const tracksEndPosition = (commonStore.trackTotalTime - commonStore.trackCurrentTime) / (baseMoveTime * scale) * snapValue;
+  const moveableTracksWidth = (tracksEndLineRef.current && viewControlsContainerRef.current &&
+    ((tracksEndLineRef.current.getBoundingClientRect().left - (viewControlsContainerRef.current.getBoundingClientRect().left + trackInitOffsetPx)) > 0))
+    ? tracksEndLineRef.current.getBoundingClientRect().left - (viewControlsContainerRef.current.getBoundingClientRect().left + trackInitOffsetPx)
+    : Infinity;
+
+  return (
+    <div className="flex flex-col gap-2 overflow-hidden" style={{ width: '80vw', height: '80vh' }}>
+      <div className="mx-auto">
+        <Text size={100}>{`${commonStore.trackPlayStartTime} ms / ${commonStore.trackTotalTime} ms`}</Text>
+      </div>
+      <div className="flex pb-2 border-b" ref={toolbarRef}>
+        <div className="flex gap-2" ref={toolbarButtonRef}>
+          <ToolTipButton disabled desc={t('Play All') + ' (Unavailable)'} icon={<Play16Regular />} />
+          <ToolTipButton desc={t('Clear All')} icon={<Delete16Regular />} onClick={() => {
+            commonStore.setTracks([]);
+            commonStore.setTrackScale(1);
+            commonStore.setTrackTotalTime(tracksMinimalTotalTime);
+            commonStore.setTrackCurrentTime(0);
+            commonStore.setTrackPlayStartTime(0);
+          }} />
+        </div>
+        <div className="grow">
+          <div className="flex flex-col ml-2 mr-2" ref={viewControlsContainerRef}>
+            <div className="relative">
+              <Tooltip content={`${commonStore.trackTotalTime} ms`} showDelay={0} hideDelay={0}
+                relationship="description">
+                <div className="border-l absolute"
+                  ref={tracksEndLineRef}
+                  style={{
+                    height: (tracksRef.current && commonStore.tracks.length > 0)
+                      ? tracksRef.current.clientHeight - arrowIconToTracks
+                      : 0,
+                    top: `${topToArrowIcon + arrowIconToTracks}px`,
+                    left: `${tracksEndPosition + trackInitOffsetPx - pixelFix}px`
+                  }} />
+              </Tooltip>
+            </div>
+            <Draggable axis="x" bounds={{
+              left: 0,
+              right: viewControlsContainerWidth - currentTimeControlWidth
+            }}
+              position={{
+                x: commonStore.trackCurrentTime / commonStore.trackTotalTime * viewControlsContainerWidth,
+                y: 0
+              }}
+              onDrag={(e, data) => {
+                setTimeout(() => {
+                  let offset = 0;
+                  if (currentTimeControlRef.current) {
+                    const match = currentTimeControlRef.current.style.transform.match(/translate\((.+)px,/);
+                    if (match)
+                      offset = parseFloat(match[1]);
+                  }
+                  const offsetTime = commonStore.trackTotalTime / viewControlsContainerWidth * offset;
+                  commonStore.setTrackCurrentTime(offsetTime);
+                }, 1);
+              }}
+            >
+              <div ref={currentTimeControlRef}
+                className={classnames('h-2 cursor-move rounded', commonStore.settings.darkMode ? 'bg-neutral-600' : 'bg-gray-700')}
+                style={{ width: currentTimeControlWidth }} />
+            </Draggable>
+            <div className={classnames(
+              'flex',
+              (playStartTimeControlPosition < 0 || playStartTimeControlPosition > viewControlsContainerWidth)
+              && 'hidden'
+            )}>
+              <Draggable axis="x" bounds={{
+                left: 0,
+                right: (playStartTimeControlRef.current)
+                  ? Math.min(viewControlsContainerWidth - playStartTimeControlRef.current.clientWidth, moveableTracksWidth)
+                  : 0
+              }}
+                grid={[snapValue, snapValue]}
+                position={{ x: playStartTimeControlPosition, y: 0 }}
+                onStart={(e, data) => {
+                  playStartTimeControlX.current = data.lastX;
+                }}
+                onStop={(e, data) => {
+                  const delta = data.lastX - playStartTimeControlX.current;
+                  let offsetTime = Math.round(Math.round(delta / snapValue * baseMoveTime * scale) / minimalMoveTime) * minimalMoveTime;
+                  offsetTime = Math.min(Math.max(
+                    offsetTime,
+                    -commonStore.trackPlayStartTime), commonStore.trackTotalTime - commonStore.trackPlayStartTime);
+                  commonStore.setTrackPlayStartTime(commonStore.trackPlayStartTime + offsetTime);
+                }}
+              >
+                <div className="relative cursor-move"
+                  ref={playStartTimeControlRef}>
+                  <ArrowAutofitWidth20Regular />
+                  <div
+                    className={classnames('border-l absolute', commonStore.settings.darkMode ? 'border-white' : 'border-gray-700')}
+                    style={{
+                      height: (tracksRef.current && commonStore.tracks.length > 0)
+                        ? tracksRef.current.clientHeight
+                        : 0,
+                      top: '50%',
+                      left: `calc(50% - ${pixelFix}px)`
+                    }} />
+                </div>
+              </Draggable>
+            </div>
+          </div>
+        </div>
+        <Tooltip content={t('Scale View')! + ': ' + commonStore.trackScale} showDelay={0} hideDelay={0}
+          relationship="description">
+          <Slider ref={toolbarSliderRef} value={commonStore.trackScale} step={scaleMin} max={scaleMax} min={scaleMin}
+            onChange={(e, data) => {
+              commonStore.setTrackScale(data.value);
+            }}
+          />
+        </Tooltip>
+      </div>
+      <div className="flex flex-col overflow-y-auto gap-1" ref={tracksRef}>
+        {commonStore.tracks.map(track =>
+          <div key={track.id} className="flex gap-2 pb-1 border-b">
+            <div className="flex gap-1 border-r h-7">
+              <ToolTipButton desc={commonStore.recordingTrackId === track.id ? t('Stop') : t('Record')}
+                disabled={commonStore.platform === 'web'}
+                icon={commonStore.recordingTrackId === track.id ? <Stop16Filled /> : <Record16Regular />}
+                size="small" shape="circular" appearance="subtle"
+                onClick={() => {
+                  flushMidiRecordingContent();
+                  commonStore.setPlayingTrackId('');
+
+                  if (commonStore.recordingTrackId === track.id) {
+                    commonStore.setRecordingTrackId('');
+                  } else {
+                    if (commonStore.activeMidiDeviceIndex === -1) {
+                      toast(t('Please select a MIDI device first'), { type: 'warning' });
+                      return;
+                    }
+
+                    dropRecordingTime = true;
+                    setSelectedTrackId(track.id);
+
+                    commonStore.setRecordingTrackId(track.id);
+                    commonStore.setRecordingContent(track.content);
+                    commonStore.setRecordingRawContent(track.rawContent.slice());
+                  }
+                }} />
+              <ToolTipButton disabled
+                desc={commonStore.playingTrackId === track.id ? t('Stop') : t('Play') + ' (Unavailable)'}
+                icon={commonStore.playingTrackId === track.id ? <Pause16Regular /> : <Play16Filled />}
+                size="small" shape="circular" appearance="subtle"
+                onClick={() => {
+                  flushMidiRecordingContent();
+                  commonStore.setRecordingTrackId('');
+
+                  if (commonStore.playingTrackId === track.id) {
+                    commonStore.setPlayingTrackId('');
+                  } else {
+                    setSelectedTrackId(track.id);
+
+                    commonStore.setPlayingTrackId(track.id);
+                  }
+                }} />
+              <ToolTipButton desc={t('Delete')} icon={<Delete16Regular />} size="small" shape="circular"
+                appearance="subtle" onClick={() => {
+                const tracks = commonStore.tracks.slice().filter(t => t.id !== track.id);
+                commonStore.setTracks(tracks);
+                refreshTracksTotalTime();
+              }} />
+            </div>
+            <div className="relative grow overflow-hidden">
+              <div className="absolute" style={{ left: -0 }}>
+                <Track
+                  id={track.id}
+                  scale={scale}
+                  right={Math.min(tracksWidth, moveableTracksWidth)}
+                  isSelected={selectedTrackId === track.id}
+                  onSelect={setSelectedTrackId}
+                />
+              </div>
+            </div>
+          </div>)}
+        <div className="flex justify-between items-center">
+          <div className="flex gap-1">
+            <Button icon={<Add16Regular />} size="small" shape="circular"
+              appearance="subtle"
+              disabled={commonStore.platform === 'web'}
+              onClick={() => {
+                commonStore.setTracks([...commonStore.tracks, {
+                  id: uuid(),
+                  mainInstrument: '',
+                  content: '',
+                  rawContent: [],
+                  offsetTime: 0,
+                  contentTime: 0
+                }]);
+              }}>
+              {t('New Track')}
+            </Button>
+            <Button icon={<ArrowUpload16Regular />} size="small" shape="circular"
+              appearance="subtle"
+              onClick={() => {
+                if (commonStore.status.status === ModelStatus.Offline && !commonStore.settings.apiUrl && commonStore.platform !== 'web') {
+                  toast(t('Please click the button in the top right corner to start the model'), { type: 'warning' });
+                  return;
+                }
+
+                OpenOpenFileDialog('*.mid').then(async filePath => {
+                  if (!filePath)
+                    return;
+
+                  let blob: Blob;
+                  if (commonStore.platform === 'web')
+                    blob = (filePath as unknown as { blob: Blob }).blob;
+                  else
+                    blob = await fetch(absPathAsset(filePath)).then(r => r.blob());
+                  const bodyForm = new FormData();
+                  bodyForm.append('file_data', blob);
+                  fetch(getServerRoot(commonStore.getCurrentModelConfig().apiParameters.apiPort) + '/midi-to-text', {
+                    method: 'POST',
+                    body: bodyForm
+                  }).then(async r => {
+                      if (r.status === 200) {
+                        const text = (await r.json()).text as string;
+                        const rawContent = text.split(' ').map(tokenToMidiMessage).filter(m => m) as MidiMessage[];
+                        const tracks = commonStore.tracks.slice();
+
+                        tracks.push({
+                          id: uuid(),
+                          mainInstrument: getMidiRawContentMainInstrument(rawContent),
+                          content: text,
+                          rawContent: rawContent,
+                          offsetTime: 0,
+                          contentTime: getMidiRawContentTime(rawContent)
+                        });
+                        commonStore.setTracks(tracks);
+                        refreshTracksTotalTime();
+                      } else {
+                        toast(r.statusText + '\n' + (await r.text()), {
+                          type: 'error'
+                        });
+                      }
+                    }
+                  ).catch(e => {
+                    toast(t('Error') + ' - ' + (e.message || e), { type: 'error', autoClose: 2500 });
+                  });
+                }).catch(e => {
+                  toast(t('Error') + ' - ' + (e.message || e), { type: 'error', autoClose: 2500 });
+                });
+              }}>
+              {t('Import MIDI')}
+            </Button>
+          </div>
+          <Text size={100}>
+            {t('Select a track to preview the content')}
+          </Text>
+        </div>
+      </div>
+      <div className="grow"></div>
+      {selectedTrack &&
+        <Card size="small" appearance="outline" style={{ minHeight: '150px', maxHeight: '200px' }}>
+          <div className="flex flex-col gap-1 overflow-hidden">
+            <Text size={100}>{`${t('Start Time')}: ${selectedTrack.offsetTime} ms`}</Text>
+            <Text size={100}>{`${t('Content Duration')}: ${selectedTrack.contentTime} ms`}</Text>
+            <div className="overflow-y-auto overflow-x-hidden" ref={contentPreviewRef}>
+              {selectedTrackId === commonStore.recordingTrackId
+                ? commonStore.recordingContent
+                : selectedTrack.content}
+            </div>
+          </div>
+        </Card>
+      }
+      {
+        commonStore.platform !== 'web' &&
+        <div className="flex gap-2 items-end mx-auto">
+          {t('Current Instrument') + ':'}
+          {InstrumentTypeNameMap.map((name, i) =>
+            <Text key={name} style={{ whiteSpace: 'nowrap' }}
+              className={commonStore.instrumentType === i ? 'text-blue-600' : ''}
+              weight={commonStore.instrumentType === i ? 'bold' : 'regular'}
+              size={commonStore.instrumentType === i ? 300 : 100}
+            >{t(name)}</Text>)}
+        </div>
+      }
+      <DialogTrigger disableButtonEnhancement>
+        <Button icon={<MusicNote220Regular />} style={{ minHeight: '32px' }} onClick={() => {
+          flushMidiRecordingContent();
+          commonStore.setRecordingTrackId('');
+          commonStore.setPlayingTrackId('');
+
+          const timestamp = [];
+          const sortedTracks = commonStore.tracks.slice().sort((a, b) => a.offsetTime - b.offsetTime);
+          for (const track of sortedTracks) {
+            timestamp.push(track.offsetTime);
+            let accContentTime = 0;
+            for (const msg of track.rawContent) {
+              if (msg.messageType === 'ElapsedTime') {
+                accContentTime += msg.value;
+                timestamp.push(track.offsetTime + accContentTime);
+              }
+            }
+          }
+          const sortedTimestamp = timestamp.slice().sort((a, b) => a - b);
+          const globalMessages: MidiMessage[] = sortedTimestamp.reduce((messages, current, i) =>
+              [...messages, {
+                messageType: 'ElapsedTime',
+                value: current - (i === 0 ? 0 : sortedTimestamp[i - 1])
+              } as MidiMessage]
+            , [] as MidiMessage[]);
+          for (const track of sortedTracks) {
+            let currentTime = track.offsetTime;
+            let accContentTime = 0;
+            for (const msg of track.rawContent) {
+              if (msg.messageType === 'ElapsedTime') {
+                accContentTime += msg.value;
+                currentTime = track.offsetTime + accContentTime;
+              } else if (msg.messageType === 'NoteOn' || msg.messageType === 'NoteOff') {
+                const insertIndex = sortedTimestamp.findIndex(t => t >= currentTime);
+                globalMessages.splice(insertIndex + 1, 0, msg);
+                sortedTimestamp.splice(insertIndex + 1, 0, 0); // placeholder
+              }
+            }
+          }
+          const result = ('<pad> ' + globalMessages.map(midiMessageToToken).join('')).trim();
+          commonStore.setCompositionSubmittedPrompt(result);
+          setPrompt(result);
+        }}>
+          {t('Save to generation area')}
+        </Button>
+      </DialogTrigger>
+    </div>
+  );
+});
+
+export default AudiotrackEditor;

frontend/src/pages/Chat.tsx

@@ -16,8 +16,11 @@ import {
   Attach16Regular,
   Delete28Regular,
   Dismiss16Regular,
+  Dismiss24Regular,
   RecordStop28Regular,
-  Save28Regular
+  SaveRegular,
+  TextAlignJustify24Regular,
+  TextAlignJustifyRotate9024Regular
 } from '@fluentui/react-icons';
 import { CopyButton } from '../components/CopyButton';
 import { ReadButton } from '../components/ReadButton';
@@ -25,10 +28,13 @@ import { toast } from 'react-toastify';
 import { WorkHeader } from '../components/WorkHeader';
 import { DialogButton } from '../components/DialogButton';
 import { OpenFileFolder, OpenOpenFileDialog, OpenSaveFileDialog } from '../../wailsjs/go/backend_golang/App';
-import { absPathAsset, bytesToReadable, getServerRoot, toastWithButton } from '../utils';
-import { PresetsButton } from './PresetsManager/PresetsButton';
+import { absPathAsset, bytesToReadable, getServerRoot, setActivePreset, toastWithButton } from '../utils';
 import { useMediaQuery } from 'usehooks-ts';
 import { botName, ConversationMessage, MessageType, userName, welcomeUuid } from '../types/chat';
+import { Labeled } from '../components/Labeled';
+import { ValuedSlider } from '../components/ValuedSlider';
+import { PresetsButton } from './PresetsManager/PresetsButton';
+import { webOpenOpenFileDialog } from '../utils/web-file-operations';
 
 let chatSseControllers: {
   [id: string]: AbortController
@@ -188,11 +194,125 @@ const ChatMessageItem: FC<{
   </div>;
 });
 
+const SidePanel: FC = observer(() => {
+  const [t] = useTranslation();
+  const mq = useMediaQuery('(min-width: 640px)');
+  const params = commonStore.chatParams;
+
+  return <div
+    className={classnames(
+      'flex flex-col gap-1 h-full flex-shrink-0 transition-width duration-300 ease-in-out',
+      commonStore.sidePanelCollapsed ? 'w-0' : (mq ? 'w-64' : 'w-full'),
+      !commonStore.sidePanelCollapsed && 'ml-1')
+    }>
+    <div className="flex m-1">
+      <div className="grow" />
+      <PresetsButton tab="Chat" size="medium" shape="circular" appearance="subtle" />
+      <Button size="medium" shape="circular" appearance="subtle" icon={<Dismiss24Regular />}
+        onClick={() => commonStore.setSidePanelCollapsed(true)}
+      />
+    </div>
+    <div className="flex flex-col gap-1 overflow-x-hidden overflow-y-auto p-1">
+      <Labeled flex breakline label={t('Max Response Token')}
+        desc={t('By default, the maximum number of tokens that can be answered in a single response, it can be changed by the user by specifying API parameters.')}
+        content={
+          <ValuedSlider value={params.maxResponseToken} min={100} max={8100}
+            step={100}
+            input
+            onChange={(e, data) => {
+              commonStore.setChatParams({
+                maxResponseToken: data.value
+              });
+            }} />
+        } />
+      <Labeled flex breakline label={t('Temperature')}
+        desc={t('Sampling temperature, it\'s like giving alcohol to a model, the higher the stronger the randomness and creativity, while the lower, the more focused and deterministic it will be.')}
+        content={
+          <ValuedSlider value={params.temperature} min={0} max={2} step={0.1}
+            input
+            onChange={(e, data) => {
+              commonStore.setChatParams({
+                temperature: data.value
+              });
+            }} />
+        } />
+      <Labeled flex breakline label={t('Top_P')}
+        desc={t('Just like feeding sedatives to the model. Consider the results of the top n% probability mass, 0.1 considers the top 10%, with higher quality but more conservative, 1 considers all results, with lower quality but more diverse.')}
+        content={
+          <ValuedSlider value={params.topP} min={0} max={1} step={0.1} input
+            onChange={(e, data) => {
+              commonStore.setChatParams({
+                topP: data.value
+              });
+            }} />
+        } />
+      <Labeled flex breakline label={t('Presence Penalty')}
+        desc={t('Positive values penalize new tokens based on whether they appear in the text so far, increasing the model\'s likelihood to talk about new topics.')}
+        content={
+          <ValuedSlider value={params.presencePenalty} min={0} max={2}
+            step={0.1} input
+            onChange={(e, data) => {
+              commonStore.setChatParams({
+                presencePenalty: data.value
+              });
+            }} />
+        } />
+      <Labeled flex breakline label={t('Frequency Penalty')}
+        desc={t('Positive values penalize new tokens based on their existing frequency in the text so far, decreasing the model\'s likelihood to repeat the same line verbatim.')}
+        content={
+          <ValuedSlider value={params.frequencyPenalty} min={0} max={2}
+            step={0.1} input
+            onChange={(e, data) => {
+              commonStore.setChatParams({
+                frequencyPenalty: data.value
+              });
+            }} />
+        } />
+    </div>
+    <div className="grow" />
+    {/*<Button*/}
+    {/*  icon={<FolderOpenVerticalRegular />}*/}
+    {/*  onClick={() => {*/}
+    {/*  }}>*/}
+    {/*  {t('Load Conversation')}*/}
+    {/*</Button>*/}
+    <Button
+      icon={<SaveRegular />}
+      onClick={() => {
+        let savedContent: string = '';
+        const isWorldModel = commonStore.getCurrentModelConfig().modelParameters.modelName.toLowerCase().includes('world');
+        const user = isWorldModel ? 'User' : 'Bob';
+        const bot = isWorldModel ? 'Assistant' : 'Alice';
+        commonStore.conversationOrder.forEach((uuid) => {
+          if (uuid === welcomeUuid)
+            return;
+          const messageItem = commonStore.conversation[uuid];
+          if (messageItem.type !== MessageType.Error) {
+            savedContent += `${messageItem.sender === userName ? user : bot}: ${messageItem.content}\n\n`;
+          }
+        });
+
+        OpenSaveFileDialog('*.txt', 'conversation.txt', savedContent).then((path) => {
+          if (path)
+            toastWithButton(t('Conversation Saved'), t('Open'), () => {
+              OpenFileFolder(path, false);
+            });
+        }).catch(e => {
+          toast(t('Error') + ' - ' + (e.message || e), { type: 'error', autoClose: 2500 });
+        });
+      }}>
+      {t('Save Conversation')}
+    </Button>
+  </div>;
+});
+
 const ChatPanel: FC = observer(() => {
   const { t } = useTranslation();
   const bodyRef = useRef<HTMLDivElement>(null);
   const inputRef = useRef<HTMLTextAreaElement>(null);
   const mq = useMediaQuery('(min-width: 640px)');
+  if (commonStore.sidePanelCollapsed === 'auto')
+    commonStore.setSidePanelCollapsed(!mq);
   const currentConfig = commonStore.getCurrentModelConfig();
   const apiParams = currentConfig.apiParameters;
   const port = apiParams.apiPort;
@@ -316,7 +436,7 @@ const ChatPanel: FC = observer(() => {
     const chatSseController = new AbortController();
     chatSseControllers[answerId] = chatSseController;
     fetchEventSource( // https://api.openai.com/v1/chat/completions || http://127.0.0.1:${port}/v1/chat/completions
-      getServerRoot(port) + '/v1/chat/completions',
+      getServerRoot(port, true) + '/v1/chat/completions',
       {
         method: 'POST',
         headers: {
@@ -327,8 +447,10 @@ const ChatPanel: FC = observer(() => {
           messages,
           stream: true,
           model: commonStore.settings.apiChatModelName, // 'gpt-3.5-turbo'
-          temperature: apiParams.temperature,
-          top_p: apiParams.topP,
+          temperature: commonStore.chatParams.temperature,
+          top_p: commonStore.chatParams.topP,
+          presence_penalty: commonStore.chatParams.presencePenalty,
+          frequency_penalty: commonStore.chatParams.frequencyPenalty,
           user_name: commonStore.activePreset?.userName || undefined,
           assistant_name: commonStore.activePreset?.assistantName || undefined,
           presystem: commonStore.activePreset?.presystem && undefined
@@ -352,6 +474,8 @@ const ChatPanel: FC = observer(() => {
             console.debug('json error', error);
             return;
           }
+          if (data.model)
+            commonStore.setLastModelName(data.model);
           if (data.choices && Array.isArray(data.choices) && data.choices.length > 0) {
             answer += data.choices[0]?.delta?.content || '';
             commonStore.conversation[answerId!].content = answer;
@@ -389,185 +513,170 @@ const ChatPanel: FC = observer(() => {
   }, []);
 
   return (
-    <div className="flex flex-col w-full grow gap-4 pt-4 overflow-hidden">
-      <div ref={bodyRef} className="grow overflow-y-scroll overflow-x-hidden pr-2">
-        {commonStore.conversationOrder.map(uuid =>
-          <ChatMessageItem key={uuid} uuid={uuid} onSubmit={onSubmit} />
-        )}
-      </div>
-      <div className={classnames('flex items-end', mq ? 'gap-2' : '')}>
-        <PresetsButton tab="Chat" size={mq ? 'large' : 'small'} shape="circular" appearance="subtle" />
-        <DialogButton tooltip={t('Clear')}
-          icon={<Delete28Regular />}
-          size={mq ? 'large' : 'small'} shape="circular" appearance="subtle" title={t('Clear')}
-          contentText={t('Are you sure you want to clear the conversation? It cannot be undone.')}
-          onConfirm={() => {
-            if (generating) {
-              for (const id in chatSseControllers) {
-                chatSseControllers[id].abort();
+    <div className="flex h-full grow pt-4 overflow-hidden">
+      <div className="relative flex flex-col w-full grow gap-4 overflow-hidden">
+        <Button className="absolute top-1 right-1" size="medium" shape="circular" appearance="subtle"
+          style={{ zIndex: 1 }}
+          icon={commonStore.sidePanelCollapsed ? <TextAlignJustify24Regular /> : <TextAlignJustifyRotate9024Regular />}
+          onClick={() => commonStore.setSidePanelCollapsed(!commonStore.sidePanelCollapsed)} />
+        <div ref={bodyRef} className="grow overflow-y-scroll overflow-x-hidden pr-2">
+          {commonStore.conversationOrder.map(uuid =>
+            <ChatMessageItem key={uuid} uuid={uuid} onSubmit={onSubmit} />
+          )}
+        </div>
+        <div className={classnames('flex items-end', mq ? 'gap-2' : '')}>
+          <DialogButton tooltip={t('Clear')}
+            icon={<Delete28Regular />}
+            size={mq ? 'large' : 'small'} shape="circular" appearance="subtle" title={t('Clear')}
+            contentText={t('Are you sure you want to clear the conversation? It cannot be undone.')}
+            onConfirm={() => {
+              if (generating) {
+                for (const id in chatSseControllers) {
+                  chatSseControllers[id].abort();
+                }
+                chatSseControllers = {};
               }
-              chatSseControllers = {};
-            }
-            commonStore.setConversation({});
-            commonStore.setConversationOrder([]);
-          }} />
-        <div className="relative flex grow">
-          <Textarea
-            ref={inputRef}
-            style={{ minWidth: 0 }}
-            className="grow"
-            resize="vertical"
-            placeholder={t('Type your message here')!}
-            value={commonStore.currentInput}
-            onChange={(e) => commonStore.setCurrentInput(e.target.value)}
-            onKeyDown={handleKeyDownOrClick}
-          />
-          <div className="absolute right-2 bottom-2">
-            {!commonStore.currentTempAttachment ?
-              <ToolTipButton
-                desc={commonStore.attachmentUploading ?
-                  t('Uploading Attachment') :
-                  t('Add An Attachment (Accepts pdf, txt)')}
-                icon={commonStore.attachmentUploading ?
-                  <ArrowClockwise16Regular className="animate-spin" />
-                  : <Attach16Regular />}
-                size="small" shape="circular" appearance="secondary"
-                onClick={() => {
-                  if (commonStore.status.status === ModelStatus.Offline && !commonStore.settings.apiUrl && commonStore.platform !== 'web') {
-                    toast(t('Please click the button in the top right corner to start the model'), { type: 'warning' });
-                    return;
-                  }
-
-                  if (commonStore.attachmentUploading)
-                    return;
-
-                  OpenOpenFileDialog('*.txt;*.pdf').then(async filePath => {
-                    if (!filePath)
+              setActivePreset(commonStore.activePreset);
+            }} />
+          <div className="relative flex grow">
+            <Textarea
+              ref={inputRef}
+              style={{ minWidth: 0 }}
+              className="grow"
+              resize="vertical"
+              placeholder={t('Type your message here')!}
+              value={commonStore.currentInput}
+              onChange={(e) => commonStore.setCurrentInput(e.target.value)}
+              onKeyDown={handleKeyDownOrClick}
+            />
+            <div className="absolute right-2 bottom-2">
+              {!commonStore.currentTempAttachment ?
+                <ToolTipButton
+                  desc={commonStore.attachmentUploading ?
+                    t('Processing Attachment') :
+                    t('Add An Attachment (Accepts pdf, txt)')}
+                  icon={commonStore.attachmentUploading ?
+                    <ArrowClockwise16Regular className="animate-spin" />
+                    : <Attach16Regular />}
+                  size="small" shape="circular" appearance="secondary"
+                  onClick={() => {
+                    if (commonStore.attachmentUploading)
                       return;
 
-                    commonStore.setAttachmentUploading(true);
-
-                    let blob: Blob;
-                    let attachmentName: string | undefined;
-                    let attachmentContent: string | undefined;
-                    if (commonStore.platform === 'web') {
-                      const webReturn = filePath as any;
-                      blob = webReturn.blob;
-                      attachmentName = blob.name;
-                      attachmentContent = webReturn.content;
-                    } else {
-                      // Both are slow. Communication between frontend and backend is slow. Use AssetServer Handler to read the file.
-                      // const blob = new Blob([atob(info.content as unknown as string)]); // await fetch(`data:application/octet-stream;base64,${info.content}`).then(r => r.blob());
-                      blob = await fetch(absPathAsset(filePath)).then(r => r.blob());
-                      attachmentName = filePath.split(/[\\/]/).pop();
-                    }
-                    if (attachmentContent) {
-                      commonStore.setCurrentTempAttachment(
-                        {
-                          name: attachmentName!,
-                          size: blob.size,
-                          content: attachmentContent
-                        });
-                      commonStore.setAttachmentUploading(false);
-                    } else {
-                      const urlPath = `/file-to-text?file_name=${attachmentName}`;
-                      const bodyForm = new FormData();
-                      bodyForm.append('file_data', blob, attachmentName);
-                      fetch(getServerRoot(port) + urlPath, {
-                        method: 'POST',
-                        body: bodyForm
-                      }).then(async r => {
-                          if (r.status === 200) {
-                            const pages = (await r.json()).pages as any[];
-                            if (pages.length === 1)
-                              attachmentContent = pages[0].page_content;
-                            else
-                              attachmentContent = pages.map((p, i) => `Page ${i + 1}:\n${p.page_content}`).join('\n\n');
-                            commonStore.setCurrentTempAttachment(
-                              {
-                                name: attachmentName!,
-                                size: blob.size,
-                                content: attachmentContent!
-                              });
-                          } else {
-                            toast(r.statusText + '\n' + (await r.text()), {
-                              type: 'error'
+                    const filterPattern = '*.txt;*.pdf';
+                    const setUploading = () => commonStore.setAttachmentUploading(true);
+                    // actually, status of web platform is always Offline
+                    if (commonStore.platform === 'web' || commonStore.status.status === ModelStatus.Offline || currentConfig.modelParameters.device === 'WebGPU') {
+                      webOpenOpenFileDialog(filterPattern, setUploading).then(webReturn => {
+                        if (webReturn.content)
+                          commonStore.setCurrentTempAttachment(
+                            {
+                              name: webReturn.blob.name,
+                              size: webReturn.blob.size,
+                              content: webReturn.content
                             });
-                          }
-                          commonStore.setAttachmentUploading(false);
-                        }
-                      ).catch(e => {
+                        else
+                          toast(t('File is empty'), {
+                            type: 'info',
+                            autoClose: 1000
+                          });
+                      }).catch(e => {
+                        toast(t('Error') + ' - ' + (e.message || e), { type: 'error', autoClose: 2500 });
+                      }).finally(() => {
                         commonStore.setAttachmentUploading(false);
+                      });
+                    } else {
+                      OpenOpenFileDialog(filterPattern).then(async filePath => {
+                        if (!filePath)
+                          return;
+
+                        setUploading();
+
+                        // Both are slow. Communication between frontend and backend is slow. Use AssetServer Handler to read the file.
+                        // const blob = new Blob([atob(info.content as unknown as string)]); // await fetch(`data:application/octet-stream;base64,${info.content}`).then(r => r.blob());
+                        const blob = await fetch(absPathAsset(filePath)).then(r => r.blob());
+                        const attachmentName = filePath.split(/[\\/]/).pop();
+                        const urlPath = `/file-to-text?file_name=${attachmentName}`;
+                        const bodyForm = new FormData();
+                        bodyForm.append('file_data', blob, attachmentName);
+                        fetch(getServerRoot(port) + urlPath, {
+                          method: 'POST',
+                          body: bodyForm
+                        }).then(async r => {
+                            if (r.status === 200) {
+                              const pages = (await r.json()).pages as any[];
+                              let attachmentContent: string;
+                              if (pages.length === 1)
+                                attachmentContent = pages[0].page_content;
+                              else
+                                attachmentContent = pages.map((p, i) => `Page ${i + 1}:\n${p.page_content}`).join('\n\n');
+                              if (attachmentContent)
+                                commonStore.setCurrentTempAttachment(
+                                  {
+                                    name: attachmentName!,
+                                    size: blob.size,
+                                    content: attachmentContent!
+                                  });
+                              else
+                                toast(t('File is empty'), {
+                                  type: 'info',
+                                  autoClose: 1000
+                                });
+                            } else {
+                              toast(r.statusText + '\n' + (await r.text()), {
+                                type: 'error'
+                              });
+                            }
+                          }
+                        ).catch(e => {
+                          toast(t('Error') + ' - ' + (e.message || e), { type: 'error', autoClose: 2500 });
+                        }).finally(() => {
+                          commonStore.setAttachmentUploading(false);
+                        });
+                      }).catch(e => {
                         toast(t('Error') + ' - ' + (e.message || e), { type: 'error', autoClose: 2500 });
                       });
                     }
-                  }).catch(e => {
-                    toast(t('Error') + ' - ' + (e.message || e), { type: 'error', autoClose: 2500 });
-                  });
-                }}
-              /> :
-              <div>
-                <ToolTipButton
-                  text={
-                    commonStore.currentTempAttachment.name.replace(
-                      new RegExp('(^[^\\.]{5})[^\\.]+'), '$1...')
-                  }
-                  desc={`${commonStore.currentTempAttachment.name} (${bytesToReadable(commonStore.currentTempAttachment.size)})`}
-                  size="small" shape="circular" appearance="secondary" />
-                <ToolTipButton desc={t('Remove Attachment')}
-                  icon={<Dismiss16Regular />}
-                  size="small" shape="circular" appearance="subtle"
-                  onClick={() => {
-                    commonStore.setCurrentTempAttachment(null);
-                  }} />
-              </div>
-            }
+                  }}
+                /> :
+                <div>
+                  <ToolTipButton
+                    text={
+                      commonStore.currentTempAttachment.name.replace(
+                        new RegExp('(^[^\\.]{5})[^\\.]+'), '$1...')
+                    }
+                    desc={`${commonStore.currentTempAttachment.name} (${bytesToReadable(commonStore.currentTempAttachment.size)})`}
+                    size="small" shape="circular" appearance="secondary" />
+                  <ToolTipButton desc={t('Remove Attachment')}
+                    icon={<Dismiss16Regular />}
+                    size="small" shape="circular" appearance="subtle"
+                    onClick={() => {
+                      commonStore.setCurrentTempAttachment(null);
+                    }} />
+                </div>
+              }
+            </div>
           </div>
+          <ToolTipButton desc={generating ? t('Stop') : t('Send')}
+            icon={generating ? <RecordStop28Regular /> : <ArrowCircleUp28Regular />}
+            size={mq ? 'large' : 'small'} shape="circular" appearance="subtle"
+            onClick={(e) => {
+              if (generating) {
+                for (const id in chatSseControllers) {
+                  chatSseControllers[id].abort();
+                  commonStore.conversation[id].type = MessageType.Error;
+                  commonStore.conversation[id].done = true;
+                }
+                chatSseControllers = {};
+                commonStore.setConversation(commonStore.conversation);
+                commonStore.setConversationOrder([...commonStore.conversationOrder]);
+              } else {
+                handleKeyDownOrClick(e);
+              }
+            }} />
         </div>
-        <ToolTipButton desc={generating ? t('Stop') : t('Send')}
-          icon={generating ? <RecordStop28Regular /> : <ArrowCircleUp28Regular />}
-          size={mq ? 'large' : 'small'} shape="circular" appearance="subtle"
-          onClick={(e) => {
-            if (generating) {
-              for (const id in chatSseControllers) {
-                chatSseControllers[id].abort();
-                commonStore.conversation[id].type = MessageType.Error;
-                commonStore.conversation[id].done = true;
-              }
-              chatSseControllers = {};
-              commonStore.setConversation(commonStore.conversation);
-              commonStore.setConversationOrder([...commonStore.conversationOrder]);
-            } else {
-              handleKeyDownOrClick(e);
-            }
-          }} />
-        <ToolTipButton desc={t('Save')}
-          icon={<Save28Regular />}
-          size={mq ? 'large' : 'small'} shape="circular" appearance="subtle"
-          onClick={() => {
-            let savedContent: string = '';
-            const isWorldModel = commonStore.getCurrentModelConfig().modelParameters.modelName.toLowerCase().includes('world');
-            const user = isWorldModel ? 'User' : 'Bob';
-            const bot = isWorldModel ? 'Assistant' : 'Alice';
-            commonStore.conversationOrder.forEach((uuid) => {
-              if (uuid === welcomeUuid)
-                return;
-              const messageItem = commonStore.conversation[uuid];
-              if (messageItem.type !== MessageType.Error) {
-                savedContent += `${messageItem.sender === userName ? user : bot}: ${messageItem.content}\n\n`;
-              }
-            });
-
-            OpenSaveFileDialog('*.txt', 'conversation.txt', savedContent).then((path) => {
-              if (path)
-                toastWithButton(t('Conversation Saved'), t('Open'), () => {
-                  OpenFileFolder(path, false);
-                });
-            }).catch(e => {
-              toast(t('Error') + ' - ' + (e.message || e), { type: 'error', autoClose: 2500 });
-            });
-          }} />
       </div>
+      <SidePanel />
     </div>
   );
 });

frontend/src/pages/Completion.tsx

@@ -29,8 +29,10 @@ const CompletionPanel: FC = observer(() => {
   };
 
   useEffect(() => {
-    if (inputRef.current)
+    if (inputRef.current) {
       inputRef.current.style.height = '100%';
+      inputRef.current.style.maxHeight = '100%';
+    }
     scrollToBottom();
   }, []);
 
@@ -80,7 +82,7 @@ const CompletionPanel: FC = observer(() => {
     let answer = '';
     completionSseController = new AbortController();
     fetchEventSource( // https://api.openai.com/v1/completions || http://127.0.0.1:${port}/v1/completions
-      getServerRoot(port) + '/v1/completions',
+      getServerRoot(port, true) + '/v1/completions',
       {
         method: 'POST',
         headers: {
@@ -112,6 +114,8 @@ const CompletionPanel: FC = observer(() => {
             console.debug('json error', error);
             return;
           }
+          if (data.model)
+            commonStore.setLastModelName(data.model);
           if (data.choices && Array.isArray(data.choices) && data.choices.length > 0) {
             answer += data.choices[0]?.text || data.choices[0]?.delta?.content || '';
             setPrompt(prompt + answer.replace(/\s+$/, '') + params.injectEnd.replaceAll('\\n', '\n'));
@@ -173,7 +177,7 @@ const CompletionPanel: FC = observer(() => {
             desc={t('By default, the maximum number of tokens that can be answered in a single response, it can be changed by the user by specifying API parameters.')}
             content={
               <ValuedSlider value={params.maxResponseToken} min={100} max={8100}
-                step={400}
+                step={100}
                 input
                 onChange={(e, data) => {
                   setParams({

frontend/src/pages/Composition.tsx

@@ -2,7 +2,7 @@ import 'html-midi-player';
 import React, { FC, useEffect, useRef } from 'react';
 import { observer } from 'mobx-react-lite';
 import { WorkHeader } from '../components/WorkHeader';
-import { Button, Checkbox, Textarea } from '@fluentui/react-components';
+import { Button, Checkbox, Dropdown, Option, Textarea } from '@fluentui/react-components';
 import { Labeled } from '../components/Labeled';
 import { ValuedSlider } from '../components/ValuedSlider';
 import { useTranslation } from 'react-i18next';
@@ -16,10 +16,19 @@ import { PlayerElement, VisualizerElement } from 'html-midi-player';
 import * as mm from '@magenta/music/esm/core.js';
 import { NoteSequence } from '@magenta/music/esm/protobuf.js';
 import { defaultCompositionPrompt } from './defaultConfigs';
-import { FileExists, OpenFileFolder, OpenSaveFileDialogBytes } from '../../wailsjs/go/backend_golang/App';
-import { getServerRoot, toastWithButton } from '../utils';
+import {
+  CloseMidiPort,
+  FileExists,
+  OpenFileFolder,
+  OpenMidiPort,
+  OpenSaveFileDialogBytes,
+  SaveFile,
+  StartFile
+} from '../../wailsjs/go/backend_golang/App';
+import { getServerRoot, getSoundFont, toastWithButton } from '../utils';
 import { CompositionParams } from '../types/composition';
 import { useMediaQuery } from 'usehooks-ts';
+import { AudiotrackButton } from './AudiotrackManager/AudiotrackButton';
 
 let compositionSseController: AbortController | null = null;
 
@@ -64,26 +73,16 @@ const CompositionPanel: FC = observer(() => {
   };
 
   const setSoundFont = async () => {
-    let soundUrl: string;
-    if (commonStore.compositionParams.useLocalSoundFont)
-      soundUrl = 'assets/sound-font';
-    else
-      soundUrl = !commonStore.settings.giteeUpdatesSource ?
-        `https://raw.githubusercontent.com/josStorer/sgm_plus/master` :
-        `https://gitee.com/josc146/sgm_plus/raw/master`;
-    const fallbackUrl = 'https://cdn.jsdelivr.net/gh/josstorer/sgm_plus';
-    await fetch(soundUrl + '/soundfont.json').then(r => {
-      if (!r.ok)
-        soundUrl = fallbackUrl;
-    }).catch(() => soundUrl = fallbackUrl);
     if (playerRef.current) {
-      playerRef.current.soundFont = soundUrl;
+      playerRef.current.soundFont = await getSoundFont();
     }
   };
 
   useEffect(() => {
-    if (inputRef.current)
+    if (inputRef.current) {
       inputRef.current.style.height = '100%';
+      inputRef.current.style.maxHeight = '100%';
+    }
     scrollToBottom();
 
     if (playerRef.current && visualizerRef.current) {
@@ -101,6 +100,38 @@ const CompositionPanel: FC = observer(() => {
     }
   }, []);
 
+  const externalPlayListener = () => {
+    const params = commonStore.compositionParams;
+    const saveAndPlay = async (midi: ArrayBuffer, path: string) => {
+      await SaveFile(path, Array.from(new Uint8Array(midi)));
+      StartFile(path);
+    };
+    if (params.externalPlay) {
+      if (params.midi) {
+        setTimeout(() => {
+          playerRef.current?.stop();
+        });
+        saveAndPlay(params.midi, './midi/last.mid').catch((e: string) => {
+          if (e.includes('being used'))
+            saveAndPlay(params.midi!, './midi/last-2.mid');
+        });
+      }
+    }
+  };
+  useEffect(() => {
+    playerRef.current?.addEventListener('start', externalPlayListener);
+    return () => {
+      playerRef.current?.removeEventListener('start', externalPlayListener);
+    };
+  }, [params.externalPlay]);
+
+  useEffect(() => {
+    if (!(commonStore.activeMidiDeviceIndex in commonStore.midiPorts)) {
+      commonStore.setActiveMidiDeviceIndex(-1);
+      CloseMidiPort();
+    }
+  }, [commonStore.midiPorts]);
+
   const generateNs = (autoPlay: boolean) => {
     fetch(getServerRoot(port) + '/text-to-midi', {
       method: 'POST',
@@ -119,7 +150,12 @@ const CompositionPanel: FC = observer(() => {
         });
         updateNs(ns);
         if (autoPlay) {
-          playerRef.current?.start();
+          if (commonStore.compositionParams.externalPlay)
+            externalPlayListener();
+          else
+            setTimeout(() => {
+              playerRef.current?.start();
+            });
         }
       });
     });
@@ -137,7 +173,7 @@ const CompositionPanel: FC = observer(() => {
     let answer = '';
     compositionSseController = new AbortController();
     fetchEventSource( // https://api.openai.com/v1/completions || http://127.0.0.1:${port}/v1/completions
-      getServerRoot(port) + '/v1/completions',
+      getServerRoot(port, true) + '/v1/completions',
       {
         method: 'POST',
         headers: {
@@ -167,6 +203,8 @@ const CompositionPanel: FC = observer(() => {
             console.debug('json error', error);
             return;
           }
+          if (data.model)
+            commonStore.setLastModelName(data.model);
           if (data.choices && Array.isArray(data.choices) && data.choices.length > 0) {
             answer += data.choices[0]?.text || data.choices[0]?.delta?.content || '';
             setPrompt(prompt + answer.replace(/\s+$/, ''));
@@ -242,27 +280,73 @@ const CompositionPanel: FC = observer(() => {
                   }} />
               } />
             <div className="grow" />
-            <Checkbox className="select-none"
-              size="large" label={t('Use Local Sound Font')} checked={params.useLocalSoundFont}
-              onChange={async (_, data) => {
-                if (data.checked) {
-                  if (!await FileExists('assets/sound-font/accordion/instrument.json')) {
-                    toast(t('Failed to load local sound font, please check if the files exist - assets/sound-font'),
-                      { type: 'warning' });
-                    return;
+            {
+              commonStore.platform !== 'web' &&
+              <Checkbox className="select-none"
+                size="large" label={t('Use Local Sound Font')} checked={params.useLocalSoundFont}
+                onChange={async (_, data) => {
+                  if (data.checked) {
+                    if (!await FileExists('assets/sound-font/accordion/instrument.json')) {
+                      toast(t('Failed to load local sound font, please check if the files exist - assets/sound-font'),
+                        { type: 'warning' });
+                      return;
+                    }
                   }
-                }
-                setParams({
-                  useLocalSoundFont: data.checked as boolean
-                });
-                setSoundFont();
-              }} />
+                  setParams({
+                    useLocalSoundFont: data.checked as boolean
+                  });
+                  setSoundFont();
+                }} />
+            }
+            {
+              commonStore.platform === 'windows' &&
+              <Checkbox className="select-none"
+                size="large" label={t('Play With External Player')} checked={params.externalPlay}
+                onChange={async (_, data) => {
+                  setParams({
+                    externalPlay: data.checked as boolean
+                  });
+                }} />
+            }
             <Checkbox className="select-none"
               size="large" label={t('Auto Play At The End')} checked={params.autoPlay} onChange={(_, data) => {
               setParams({
                 autoPlay: data.checked as boolean
               });
             }} />
+            <Labeled flex breakline label={t('MIDI Input')}
+              desc={t('Select the MIDI input device to be used.')}
+              content={
+                <div className="flex flex-col gap-1">
+                  {
+                    commonStore.platform !== 'web' &&
+                    <Dropdown style={{ minWidth: 0 }}
+                      value={(commonStore.activeMidiDeviceIndex === -1 || !(commonStore.activeMidiDeviceIndex in commonStore.midiPorts))
+                        ? t('None')!
+                        : commonStore.midiPorts[commonStore.activeMidiDeviceIndex].name}
+                      selectedOptions={[commonStore.activeMidiDeviceIndex.toString()]}
+                      onOptionSelect={(_, data) => {
+                        if (data.optionValue) {
+                          const index = Number(data.optionValue);
+                          let action = (index === -1)
+                            ? () => CloseMidiPort()
+                            : () => OpenMidiPort(index);
+                          action().then(() => {
+                            commonStore.setActiveMidiDeviceIndex(index);
+                          }).catch((e) => {
+                            toast(t('Error') + ' - ' + (e.message || e), { type: 'error', autoClose: 2500 });
+                          });
+                        }
+                      }}>
+                      <Option value={'-1'}>{t('None')!}</Option>
+                      {commonStore.midiPorts.map((p, i) =>
+                        <Option key={i} value={i.toString()}>{p.name}</Option>)
+                      }
+                    </Dropdown>
+                  }
+                  <AudiotrackButton setPrompt={setPrompt} />
+                </div>
+              } />
           </div>
           <div className="flex justify-between gap-2">
             <ToolTipButton desc={t('Regenerate')} icon={<ArrowSync20Regular />} onClick={() => {

frontend/src/pages/Configs.tsx

@@ -8,12 +8,13 @@ import {
   Input,
   Label,
   Option,
+  PresenceBadge,
   Select,
   Switch,
   Text
 } from '@fluentui/react-components';
 import { AddCircle20Regular, DataUsageSettings20Regular, Delete20Regular, Save20Regular } from '@fluentui/react-icons';
-import React, { FC, useEffect, useRef } from 'react';
+import React, { FC, useCallback, useEffect, useRef } from 'react';
 import { Section } from '../components/Section';
 import { Labeled } from '../components/Labeled';
 import { ToolTipButton } from '../components/ToolTipButton';
@@ -26,15 +27,39 @@ import { Page } from '../components/Page';
 import { useNavigate } from 'react-router';
 import { RunButton } from '../components/RunButton';
 import { updateConfig } from '../apis';
-import { ConvertModel, ConvertSafetensors, FileExists, GetPyError } from '../../wailsjs/go/backend_golang/App';
-import { checkDependencies, getStrategy } from '../utils';
+import { getStrategy } from '../utils';
 import { useTranslation } from 'react-i18next';
-import { WindowShow } from '../../wailsjs/runtime';
 import strategyImg from '../assets/images/strategy.jpg';
 import strategyZhImg from '../assets/images/strategy_zh.jpg';
 import { ResetConfigsButton } from '../components/ResetConfigsButton';
 import { useMediaQuery } from 'usehooks-ts';
 import { ApiParameters, Device, ModelParameters, Precision } from '../types/configs';
+import { convertModel, convertToGGML, convertToSt } from '../utils/convert-model';
+
+const ConfigSelector: FC<{
+  selectedIndex: number,
+  updateSelectedIndex: (i: number) => void
+}> = observer(({ selectedIndex, updateSelectedIndex }) => {
+  return (
+    <Dropdown style={{ minWidth: 0 }} className="grow" value={commonStore.modelConfigs[selectedIndex].name}
+      selectedOptions={[selectedIndex.toString()]}
+      onOptionSelect={(_, data) => {
+        if (data.optionValue) {
+          updateSelectedIndex(Number(data.optionValue));
+        }
+      }}>
+      {commonStore.modelConfigs.map((config, index) => <Option key={index} value={index.toString()}
+          text={config.name}>
+          <div className="flex justify-between grow">
+            {config.name}
+            {commonStore.modelSourceList.find(item => item.name === config.modelParameters.modelName)?.isComplete
+              && <PresenceBadge status="available" />}
+          </div>
+        </Option>
+      )}
+    </Dropdown>
+  );
+});
 
 const Configs: FC = observer(() => {
   const { t } = useTranslation();
@@ -51,13 +76,13 @@ const Configs: FC = observer(() => {
       (advancedHeaderRef.current.firstElementChild as HTMLElement).style.padding = '0';
   }, []);
 
-  const updateSelectedIndex = (newIndex: number) => {
+  const updateSelectedIndex = useCallback((newIndex: number) => {
     setSelectedIndex(newIndex);
     setSelectedConfig(commonStore.modelConfigs[newIndex]);
 
     // if you don't want to update the config used by the current startup in real time, comment out this line
     commonStore.setCurrentConfigIndex(newIndex);
-  };
+  }, []);
 
   const setSelectedConfigName = (newName: string) => {
     setSelectedConfig({ ...selectedConfig, name: newName });
@@ -97,17 +122,7 @@ const Configs: FC = observer(() => {
     <Page title={t('Configs')} content={
       <div className="flex flex-col gap-2 overflow-hidden">
         <div className="flex gap-2 items-center">
-          <Dropdown style={{ minWidth: 0 }} className="grow" value={commonStore.modelConfigs[selectedIndex].name}
-            selectedOptions={[selectedIndex.toString()]}
-            onOptionSelect={(_, data) => {
-              if (data.optionValue) {
-                updateSelectedIndex(Number(data.optionValue));
-              }
-            }}>
-            {commonStore.modelConfigs.map((config, index) =>
-              <Option key={index} value={index.toString()}>{config.name}</Option>
-            )}
-          </Dropdown>
+          <ConfigSelector selectedIndex={selectedIndex} updateSelectedIndex={updateSelectedIndex} />
           <ToolTipButton desc={t('New Config')} icon={<AddCircle20Regular />} onClick={() => {
             commonStore.createModelConfig();
             updateSelectedIndex(commonStore.modelConfigs.length - 1);
@@ -150,7 +165,7 @@ const Configs: FC = observer(() => {
                   desc={t('By default, the maximum number of tokens that can be answered in a single response, it can be changed by the user by specifying API parameters.')}
                   content={
                     <ValuedSlider value={selectedConfig.apiParameters.maxResponseToken} min={100} max={8100}
-                      step={400}
+                      step={100}
                       input
                       onChange={(e, data) => {
                         setSelectedConfigApiParams({
@@ -231,82 +246,17 @@ const Configs: FC = observer(() => {
                   </div>
                 } />
                 {
-                  selectedConfig.modelParameters.device !== 'WebGPU' ?
-                    <ToolTipButton text={t('Convert')}
-                      desc={t('Convert model with these configs. Using a converted model will greatly improve the loading speed, but model parameters of the converted model cannot be modified.')}
-                      onClick={async () => {
-                        if (commonStore.platform === 'darwin') {
-                          toast(t('MacOS is not yet supported for performing this operation, please do it manually.') + ' (backend-python/convert_model.py)', { type: 'info' });
-                          return;
-                        } else if (commonStore.platform === 'linux') {
-                          toast(t('Linux is not yet supported for performing this operation, please do it manually.') + ' (backend-python/convert_model.py)', { type: 'info' });
-                          return;
-                        }
-
-                        const ok = await checkDependencies(navigate);
-                        if (!ok)
-                          return;
-
-                        const modelPath = `${commonStore.settings.customModelsPath}/${selectedConfig.modelParameters.modelName}`;
-                        if (await FileExists(modelPath)) {
-                          const strategy = getStrategy(selectedConfig);
-                          const newModelPath = modelPath + '-' + strategy.replace(/[:> *+]/g, '-');
-                          toast(t('Start Converting'), { autoClose: 1000, type: 'info' });
-                          ConvertModel(commonStore.settings.customPythonPath, modelPath, strategy, newModelPath).then(async () => {
-                            if (!await FileExists(newModelPath + '.pth')) {
-                              toast(t('Convert Failed') + ' - ' + await GetPyError(), { type: 'error' });
-                            } else {
-                              toast(`${t('Convert Success')} - ${newModelPath}`, { type: 'success' });
-                            }
-                          }).catch(e => {
-                            const errMsg = e.message || e;
-                            if (errMsg.includes('path contains space'))
-                              toast(`${t('Convert Failed')} - ${t('File Path Cannot Contain Space')}`, { type: 'error' });
-                            else
-                              toast(`${t('Convert Failed')} - ${e.message || e}`, { type: 'error' });
-                          });
-                          setTimeout(WindowShow, 1000);
-                        } else {
-                          toast(`${t('Model Not Found')} - ${modelPath}`, { type: 'error' });
-                        }
-                      }} /> :
-                    <ToolTipButton text={t('Convert To Safe Tensors Format')}
+                  !selectedConfig.modelParameters.device.startsWith('WebGPU') ?
+                    (selectedConfig.modelParameters.device !== 'CPU (rwkv.cpp)' ?
+                      <ToolTipButton text={t('Convert')}
+                        desc={t('Convert model with these configs. Using a converted model will greatly improve the loading speed, but model parameters of the converted model cannot be modified.')}
+                        onClick={() => convertModel(selectedConfig, navigate)} /> :
+                      <ToolTipButton text={t('Convert To GGML Format')}
+                        desc=""
+                        onClick={() => convertToGGML(selectedConfig, navigate)} />)
+                    : <ToolTipButton text={t('Convert To Safe Tensors Format')}
                       desc=""
-                      onClick={async () => {
-                        if (commonStore.platform === 'darwin') {
-                          toast(t('MacOS is not yet supported for performing this operation, please do it manually.') + ' (backend-python/convert_safetensors.py)', { type: 'info' });
-                          return;
-                        } else if (commonStore.platform === 'linux') {
-                          toast(t('Linux is not yet supported for performing this operation, please do it manually.') + ' (backend-python/convert_safetensors.py)', { type: 'info' });
-                          return;
-                        }
-
-                        const ok = await checkDependencies(navigate);
-                        if (!ok)
-                          return;
-
-                        const modelPath = `${commonStore.settings.customModelsPath}/${selectedConfig.modelParameters.modelName}`;
-                        if (await FileExists(modelPath)) {
-                          toast(t('Start Converting'), { autoClose: 1000, type: 'info' });
-                          const newModelPath = modelPath.replace(/\.pth$/, '.st');
-                          ConvertSafetensors(commonStore.settings.customPythonPath, modelPath, newModelPath).then(async () => {
-                            if (!await FileExists(newModelPath)) {
-                              toast(t('Convert Failed') + ' - ' + await GetPyError(), { type: 'error' });
-                            } else {
-                              toast(`${t('Convert Success')} - ${newModelPath}`, { type: 'success' });
-                            }
-                          }).catch(e => {
-                            const errMsg = e.message || e;
-                            if (errMsg.includes('path contains space'))
-                              toast(`${t('Convert Failed')} - ${t('File Path Cannot Contain Space')}`, { type: 'error' });
-                            else
-                              toast(`${t('Convert Failed')} - ${e.message || e}`, { type: 'error' });
-                          });
-                          setTimeout(WindowShow, 1000);
-                        } else {
-                          toast(`${t('Model Not Found')} - ${modelPath}`, { type: 'error' });
-                        }
-                      }} />
+                      onClick={() => convertToSt(selectedConfig, navigate)} />
                 }
                 <Labeled label={t('Strategy')} content={
                   <Dropdown style={{ minWidth: 0 }} className="grow" value={t(selectedConfig.modelParameters.device)!}
@@ -319,18 +269,21 @@ const Configs: FC = observer(() => {
                       }
                     }}>
                     <Option value="CPU">CPU</Option>
+                    <Option value="CPU (rwkv.cpp)">{t('CPU (rwkv.cpp, Faster)')!}</Option>
                     {commonStore.platform === 'darwin' && <Option value="MPS">MPS</Option>}
                     <Option value="CUDA">CUDA</Option>
                     <Option value="CUDA-Beta">{t('CUDA (Beta, Faster)')!}</Option>
                     <Option value="WebGPU">WebGPU</Option>
+                    <Option value="WebGPU (Python)">WebGPU (Python)</Option>
                     <Option value="Custom">{t('Custom')!}</Option>
                   </Dropdown>
                 } />
                 {
                   selectedConfig.modelParameters.device !== 'Custom' && <Labeled label={t('Precision')}
-                    desc={t('int8 uses less VRAM, but has slightly lower quality. fp16 has higher quality, and fp32 has the best quality.')}
+                    desc={t('int8 uses less VRAM, but has slightly lower quality. fp16 has higher quality.')}
                     content={
-                      <Dropdown style={{ minWidth: 0 }} className="grow"
+                      <Dropdown
+                        style={{ minWidth: 0 }} className="grow"
                         value={selectedConfig.modelParameters.precision}
                         selectedOptions={[selectedConfig.modelParameters.precision]}
                         onOptionSelect={(_, data) => {
@@ -340,19 +293,23 @@ const Configs: FC = observer(() => {
                             });
                           }
                         }}>
-                        <Option>fp16</Option>
-                        <Option>int8</Option>
-                        {selectedConfig.modelParameters.device !== 'WebGPU' && <Option>fp32</Option>}
+                        {selectedConfig.modelParameters.device !== 'CPU' && selectedConfig.modelParameters.device !== 'MPS' &&
+                          <Option>fp16</Option>}
+                        {selectedConfig.modelParameters.device !== 'CPU (rwkv.cpp)' && <Option>int8</Option>}
+                        {selectedConfig.modelParameters.device.startsWith('WebGPU') && <Option>nf4</Option>}
+                        {selectedConfig.modelParameters.device !== 'CPU (rwkv.cpp)' && !selectedConfig.modelParameters.device.startsWith('WebGPU') &&
+                          <Option>fp32</Option>}
+                        {selectedConfig.modelParameters.device === 'CPU (rwkv.cpp)' && <Option>Q5_1</Option>}
                       </Dropdown>
                     } />
                 }
                 {
-                  selectedConfig.modelParameters.device.includes('CUDA') &&
+                  selectedConfig.modelParameters.device.startsWith('CUDA') &&
                   <Labeled label={t('Current Strategy')}
                     content={<Text> {getStrategy(selectedConfig)} </Text>} />
                 }
                 {
-                  selectedConfig.modelParameters.device.includes('CUDA') &&
+                  selectedConfig.modelParameters.device.startsWith('CUDA') &&
                   <Labeled label={t('Stored Layers')}
                     desc={t('Number of the neural network layers loaded into VRAM, the more you load, the faster the speed, but it consumes more VRAM. (If your VRAM is not enough, it will fail to load)')}
                     content={
@@ -365,7 +322,7 @@ const Configs: FC = observer(() => {
                         }} />
                     } />
                 }
-                {selectedConfig.modelParameters.device.includes('CUDA') && <div />}
+                {selectedConfig.modelParameters.device.startsWith('CUDA') && <div />}
                 {
                   displayStrategyImg &&
                   <img style={{ width: '80vh', height: 'auto', zIndex: 100 }}
@@ -390,7 +347,7 @@ const Configs: FC = observer(() => {
                 }
                 {selectedConfig.modelParameters.device === 'Custom' && <div />}
                 {
-                  (selectedConfig.modelParameters.device.includes('CUDA') || selectedConfig.modelParameters.device === 'Custom') &&
+                  (selectedConfig.modelParameters.device.startsWith('CUDA') || selectedConfig.modelParameters.device === 'Custom') &&
                   <Labeled label={t('Use Custom CUDA kernel to Accelerate')}
                     desc={t('Enabling this option can greatly improve inference speed and save some VRAM, but there may be compatibility issues (output garbled). If it fails to start, please turn off this option, or try to upgrade your gpu driver.')}
                     content={
@@ -423,7 +380,7 @@ const Configs: FC = observer(() => {
                                 });
                               }} />
                             <Input className="grow"
-                              placeholder={t('Tokenizer Path (e.g. backend-python/rwkv_pip/20B_tokenizer.json)')!}
+                              placeholder={t('Tokenizer Path (e.g. backend-python/rwkv_pip/20B_tokenizer.json or rwkv_vocab_v20230424.txt)')!}
                               value={selectedConfig.modelParameters.customTokenizer}
                               onChange={(e, data) => {
                                 setSelectedConfigModelParams({
@@ -439,6 +396,7 @@ const Configs: FC = observer(() => {
               </div>
             }
           />
+          {mq && <div style={{ minHeight: '30px' }} />}
         </div>
         <div className="flex flex-row-reverse sm:fixed bottom-2 right-2">
           <div className="flex gap-2">

frontend/src/pages/Models.tsx

@@ -1,5 +1,7 @@
-import React, { FC } from 'react';
+import React, { FC, useEffect, useState } from 'react';
 import {
+  Button,
+  Checkbox,
   createTableColumn,
   DataGrid,
   DataGridBody,
@@ -19,7 +21,7 @@ import commonStore from '../stores/commonStore';
 import { BrowserOpenURL } from '../../wailsjs/runtime';
 import { AddToDownloadList, OpenFileFolder } from '../../wailsjs/go/backend_golang/App';
 import { Page } from '../components/Page';
-import { bytesToGb, refreshModels, saveConfigs, toastWithButton } from '../utils';
+import { bytesToGb, getHfDownloadUrl, refreshModels, saveConfigs, toastWithButton } from '../utils';
 import { useTranslation } from 'react-i18next';
 import { useNavigate } from 'react-router';
 import { ModelSourceItem } from '../types/models';
@@ -139,7 +141,7 @@ const columns: TableColumnDefinition<ModelSourceItem>[] = [
                     navigate({ pathname: '/downloads' });
                   },
                   { autoClose: 3000 });
-                AddToDownloadList(`${commonStore.settings.customModelsPath}/${item.name}`, item.downloadUrl!);
+                AddToDownloadList(`${commonStore.settings.customModelsPath}/${item.name}`, getHfDownloadUrl(item.downloadUrl!));
               }} />}
             {item.url && <ToolTipButton desc={t('Open Url')} icon={<Open20Regular />} onClick={() => {
               BrowserOpenURL(item.url!);
@@ -153,6 +155,22 @@ const columns: TableColumnDefinition<ModelSourceItem>[] = [
 
 const Models: FC = observer(() => {
   const { t } = useTranslation();
+  const [tags, setTags] = useState<Array<string>>([]);
+  const [modelSourceList, setModelSourceList] = useState<ModelSourceItem[]>(commonStore.modelSourceList);
+
+  useEffect(() => {
+    setTags(Array.from(new Set(
+      [...commonStore.modelSourceList.map(item => item.tags || []).flat()
+      .filter(i => !i.includes('Other') && !i.includes('Local'))
+        , 'Other', 'Local'])));
+  }, [commonStore.modelSourceList]);
+
+  useEffect(() => {
+    if (commonStore.activeModelListTags.length === 0)
+      setModelSourceList(commonStore.modelSourceList);
+    else
+      setModelSourceList(commonStore.modelSourceList.filter(item => commonStore.activeModelListTags.some(tag => item.tags?.includes(tag))));
+  }, [commonStore.modelSourceList, commonStore.activeModelListTags]);
 
   return (
     <Page title={t('Models')} content={
@@ -160,10 +178,21 @@ const Models: FC = observer(() => {
         <div className="flex flex-col gap-1">
           <div className="flex justify-between items-center">
             <Text weight="medium">{t('Model Source Manifest List')}</Text>
-            <ToolTipButton desc={t('Refresh')} icon={<ArrowClockwise20Regular />} onClick={() => {
-              refreshModels(false);
-              saveConfigs();
-            }} />
+            <div className="flex">
+              {commonStore.settings.language === 'zh' &&
+                <Checkbox className="select-none"
+                  size="large" label={t('Use Hugging Face Mirror')}
+                  checked={commonStore.settings.useHfMirror}
+                  onChange={(_, data) => {
+                    commonStore.setSettings({
+                      useHfMirror: data.checked as boolean
+                    });
+                  }} />}
+              <ToolTipButton desc={t('Refresh')} icon={<ArrowClockwise20Regular />} onClick={() => {
+                refreshModels(false);
+                saveConfigs();
+              }} />
+            </div>
           </div>
           <Text size={100}>
             {t('Provide JSON file URLs for the models manifest. Separate URLs with semicolons. The "models" field in JSON files will be parsed into the following table.')}
@@ -172,9 +201,24 @@ const Models: FC = observer(() => {
             value={commonStore.modelSourceManifestList}
             onChange={(e, data) => commonStore.setModelSourceManifestList(data.value)} />
         </div>
+        <div className="flex gap-2 flex-wrap overflow-y-auto" style={{ minHeight: '88px' }}>
+          {tags.map(tag =>
+            <div key={tag} className="mt-auto">
+              <Button
+                appearance={commonStore.activeModelListTags.includes(tag) ? 'primary' : 'secondary'} onClick={
+                () => {
+                  if (commonStore.activeModelListTags.includes(tag))
+                    commonStore.setActiveModelListTags(commonStore.activeModelListTags.filter(t => t !== tag));
+                  else
+                    commonStore.setActiveModelListTags([...commonStore.activeModelListTags, tag]);
+                }
+              }>{t(tag)}</Button>
+            </div>)
+          }
+        </div>
         <div className="flex grow overflow-hidden">
           <DataGrid
-            items={commonStore.modelSourceList}
+            items={modelSourceList}
             columns={columns}
             sortable={true}
             defaultSortState={{ sortColumn: 'actions', sortDirection: 'ascending' }}

frontend/src/pages/PresetsManager/MessagesEditor.tsx

@@ -108,7 +108,8 @@ const MessagesEditor: FC = observer(() => {
                             style={{ borderTopRightRadius: 0, borderBottomRightRadius: 0 }}>
                             <ReOrderDotsVertical20Regular />
                           </Card>
-                          <Dropdown style={{ minWidth: 0, borderRadius: 0 }} listbox={{ style: { minWidth: 0 } }}
+                          <Dropdown style={{ minWidth: 0, borderRadius: 0 }}
+                            listbox={{ style: { minWidth: 'fit-content' } }}
                             value={t(item.role)!}
                             selectedOptions={[item.role]}
                             onOptionSelect={(_, data) => {

frontend/src/pages/PresetsManager/PresetsButton.tsx

@@ -33,10 +33,8 @@ import { observer } from 'mobx-react-lite';
 import { ClipboardGetText, ClipboardSetText } from '../../../wailsjs/runtime';
 import { toast } from 'react-toastify';
 import { CustomToastContainer } from '../../components/CustomToastContainer';
-import { v4 as uuid } from 'uuid';
-import { absPathAsset } from '../../utils';
+import { absPathAsset, setActivePreset } from '../../utils';
 import { Preset, PresetsNavigationItem } from '../../types/presets';
-import { botName, Conversation, MessageType, userName } from '../../types/chat';
 import { LazyImportComponent } from '../../components/LazyImportComponent';
 
 const defaultPreset: Preset = {
@@ -52,35 +50,17 @@ const defaultPreset: Preset = {
   prompt: '',
   stop: '',
   injectStart: '',
-  injectEnd: ''
+  injectEnd: '',
+  presystem: true,
+  userName: '',
+  assistantName: ''
 };
 
 const MessagesEditor = lazy(() => import('./MessagesEditor'));
 
-const setActivePreset = (preset: Preset) => {
-  commonStore.setActivePreset(preset);
-  //TODO if (preset.displayPresetMessages) {
-  const conversation: Conversation = {};
-  const conversationOrder: string[] = [];
-  for (const message of preset.messages) {
-    const newUuid = uuid();
-    conversationOrder.push(newUuid);
-    conversation[newUuid] = {
-      sender: message.role === 'user' ? userName : botName,
-      type: MessageType.Normal,
-      color: message.role === 'user' ? 'brand' : 'colorful',
-      time: new Date().toISOString(),
-      content: message.content,
-      side: message.role === 'user' ? 'right' : 'left',
-      done: true
-    };
-  }
-  commonStore.setConversation(conversation);
-  commonStore.setConversationOrder(conversationOrder);
-  //}
-};
-
-const PresetCardFrame: FC<PropsWithChildren & { onClick?: () => void }> = (props) => {
+const PresetCardFrame: FC<PropsWithChildren & {
+  onClick?: React.MouseEventHandler<HTMLButtonElement>
+}> = (props) => {
   return <Button
     className="flex flex-col gap-1 w-32 h-56 break-all"
     style={{ minWidth: 0, borderRadius: '0.75rem', justifyContent: 'unset' }}
@@ -103,7 +83,10 @@ const PresetCard: FC<{
 }) => {
   const { t } = useTranslation();
 
-  return <PresetCardFrame onClick={onClick}>
+  return <PresetCardFrame onClick={(e) => {
+    if (onClick && e.currentTarget.contains(e.target as Node))
+      onClick();
+  }}>
     <img src={absPathAsset(avatarImg)} className="rounded-xl select-none ml-auto mr-auto h-28" />
     <Text size={400}>{name}</Text>
     <Text size={200} style={{
@@ -116,7 +99,8 @@ const PresetCard: FC<{
       {editable ?
         <ChatPresetEditor presetIndex={presetIndex} triggerButton={
           <ToolTipButton size="small" appearance="transparent" desc={t('Edit')} icon={<Edit20Regular />}
-            onClick={() => {
+            onClick={(e) => {
+              e.stopPropagation();
               commonStore.setEditingPreset({ ...commonStore.presets[presetIndex] });
             }} />
         } />
@@ -371,7 +355,9 @@ const ChatPresets: FC = observer(() => {
   </div>;
 });
 
-const pages: { [label: string]: PresetsNavigationItem } = {
+const pages: {
+  [label: string]: PresetsNavigationItem
+} = {
   Chat: {
     icon: <Chat20Regular />,
     element: <ChatPresets />
@@ -386,7 +372,9 @@ const pages: { [label: string]: PresetsNavigationItem } = {
   }
 };
 
-const PresetsManager: FC<{ initTab: string }> = ({ initTab }) => {
+const PresetsManager: FC<{
+  initTab: string
+}> = ({ initTab }) => {
   const { t } = useTranslation();
   const [tab, setTab] = useState(initTab);
 

frontend/src/pages/Settings.tsx

@@ -23,7 +23,7 @@ export const GeneralSettings: FC = observer(() => {
 
   return <div className="flex flex-col gap-2">
     <Labeled label={t('Language')} flex spaceBetween content={
-      <Dropdown style={{ minWidth: 0 }} listbox={{ style: { minWidth: 0 } }}
+      <Dropdown style={{ minWidth: 0 }} listbox={{ style: { minWidth: 'fit-content' } }}
         value={Languages[commonStore.settings.language]}
         selectedOptions={[commonStore.settings.language]}
         onOptionSelect={(_, data) => {
@@ -43,7 +43,7 @@ export const GeneralSettings: FC = observer(() => {
     {
       commonStore.platform === 'windows' &&
       <Labeled label={t('DPI Scaling')} flex spaceBetween content={
-        <Dropdown style={{ minWidth: 0 }} listbox={{ style: { minWidth: 0 } }}
+        <Dropdown style={{ minWidth: 0 }} listbox={{ style: { minWidth: 'fit-content' } }}
           value={commonStore.settings.dpiScaling + '%'}
           selectedOptions={[commonStore.settings.dpiScaling.toString()]}
           onOptionSelect={(_, data) => {
@@ -89,7 +89,7 @@ export const AdvancedGeneralSettings: FC = observer(() => {
                 apiUrl: data.value
               });
             }} />
-          <Dropdown style={{ minWidth: '33px' }} listbox={{ style: { minWidth: 0 } }}
+          <Dropdown style={{ minWidth: '33px' }} listbox={{ style: { minWidth: 'fit-content' } }}
             value="..." selectedOptions={[]} expandIcon={null}
             onOptionSelect={(_, data) => {
               commonStore.setSettings({
@@ -102,11 +102,21 @@ export const AdvancedGeneralSettings: FC = observer(() => {
                   });
                 if (commonStore.settings.apiCompletionModelName === 'rwkv')
                   commonStore.setSettings({
-                    apiCompletionModelName: 'text-davinci-003'
+                    apiCompletionModelName: 'gpt-3.5-turbo-instruct'
+                  });
+              } else if (data.optionText === 'RWKV') {
+                if (commonStore.settings.apiChatModelName === 'gpt-3.5-turbo')
+                  commonStore.setSettings({
+                    apiChatModelName: 'rwkv'
+                  });
+                if (commonStore.settings.apiCompletionModelName === 'gpt-3.5-turbo-instruct' || commonStore.settings.apiCompletionModelName === 'text-davinci-003')
+                  commonStore.setSettings({
+                    apiCompletionModelName: 'rwkv'
                   });
               }
             }}>
             <Option value="">{t('Localhost')!}</Option>
+            <Option value="https://rwkv.ai-creator.net/chntuned">RWKV</Option>
             <Option value="https://api.openai.com">OpenAI</Option>
           </Dropdown>
         </div>
@@ -130,7 +140,7 @@ export const AdvancedGeneralSettings: FC = observer(() => {
                 apiChatModelName: data.value
               });
             }} />
-          <Dropdown style={{ minWidth: '33px' }} listbox={{ style: { minWidth: 0 } }}
+          <Dropdown style={{ minWidth: '33px' }} listbox={{ style: { minWidth: 'fit-content' } }}
             value="..." selectedOptions={[]} expandIcon={null}
             onOptionSelect={(_, data) => {
               if (data.optionValue) {
@@ -140,7 +150,7 @@ export const AdvancedGeneralSettings: FC = observer(() => {
               }
             }}>
             {
-              ['rwkv', 'gpt-4', 'gpt-4-0613', 'gpt-4-32k', 'gpt-4-32k-0613', 'gpt-3.5-turbo', 'gpt-3.5-turbo-0613', 'gpt-3.5-turbo-16k', 'gpt-3.5-turbo-16k-0613']
+              ['rwkv', 'gpt-4-1106-preview', 'gpt-4', 'gpt-4-32k', 'gpt-4-0613', 'gpt-4-32k-0613', 'gpt-3.5-turbo-1106', 'gpt-3.5-turbo', 'gpt-3.5-turbo-16k']
               .map((v, i) =>
                 <Option key={i} value={v}>{v}</Option>
               )
@@ -158,7 +168,7 @@ export const AdvancedGeneralSettings: FC = observer(() => {
                 apiCompletionModelName: data.value
               });
             }} />
-          <Dropdown style={{ minWidth: '33px' }} listbox={{ style: { minWidth: 0 } }}
+          <Dropdown style={{ minWidth: '33px' }} listbox={{ style: { minWidth: 'fit-content', minHeight: 0 } }}
             value="..." selectedOptions={[]} expandIcon={null}
             onOptionSelect={(_, data) => {
               if (data.optionValue) {
@@ -168,7 +178,7 @@ export const AdvancedGeneralSettings: FC = observer(() => {
               }
             }}>
             {
-              ['rwkv', 'text-davinci-003', 'text-davinci-002', 'text-curie-001', 'text-babbage-001', 'text-ada-001']
+              ['rwkv', 'gpt-3.5-turbo-instruct', 'text-davinci-003', 'text-davinci-002', 'code-davinci-002', 'text-curie-001', 'text-babbage-001', 'text-ada-001']
               .map((v, i) =>
                 <Option key={i} value={v}>{v}</Option>
               )
@@ -176,6 +186,16 @@ export const AdvancedGeneralSettings: FC = observer(() => {
           </Dropdown>
         </div>
       } />
+    <Labeled label={t('Core API URL')}
+      desc={t('Override core API URL(/chat/completions and /completions). If you don\'t know what this is, leave it blank.')}
+      content={
+        <Input style={{ minWidth: 0 }} className="grow" value={commonStore.settings.coreApiUrl}
+          onChange={(e, data) => {
+            commonStore.setSettings({
+              coreApiUrl: data.value
+            });
+          }} />
+      } />
   </div>;
 });
 

frontend/src/pages/Train.tsx

@@ -66,6 +66,11 @@ const parseLossData = (data: string) => {
   const loss = parseFloat(lastMatch[8]);
   commonStore.setChartTitle(`Epoch ${epoch}: ${lastMatch[2]} - ${lastMatch[3]}/${lastMatch[4]} - ${lastMatch[5]}/${lastMatch[6]} - ${lastMatch[7]} Loss=${loss}`);
   addLossDataToChart(epoch, loss);
+  if (loss > 5)
+    toast(t('Loss is too high, please check the training data, and ensure your gpu driver is up to date.'), {
+      type: 'warning',
+      toastId: 'train_loss_high'
+    });
   return true;
 };
 
@@ -129,11 +134,13 @@ const errorsMap = Object.entries({
   'python3 ./finetune/lora/train.py': 'Memory is not enough, try to increase the virtual memory (Swap of WSL) or use a smaller base model.',
   'cuda out of memory': 'VRAM is not enough',
   'valueerror: high <= 0': 'Training data is not enough, reduce context length or add more data for training',
-  '+= \'+ptx\'': 'You are using WSL 1 for training, please upgrade to WSL 2. e.g. Run "wsl --set-version Ubuntu-22.04 2"',
+  '+= \'+ptx\'': 'Can not find an Nvidia GPU. Perhaps the gpu driver of windows is too old, or you are using WSL 1 for training, please upgrade to WSL 2. e.g. Run "wsl --set-version Ubuntu-22.04 2"',
   'size mismatch for blocks': 'Size mismatch for blocks. You are attempting to continue training from the LoRA model, but it does not match the base model. Please set LoRA model to None.',
   'cuda_home environment variable is not set': 'Matched CUDA is not installed',
   'unsupported gpu architecture': 'Matched CUDA is not installed',
-  'error building extension \'fused_adam\'': 'Matched CUDA is not installed'
+  'error building extension \'fused_adam\'': 'Matched CUDA is not installed',
+  'rwkv{version} is not supported': 'This version of RWKV is not supported yet.',
+  'modelinfo is invalid': 'Failed to load model, try to increase the virtual memory (Swap of WSL) or use a smaller base model.'
 });
 
 export const wslHandler = (data: string) => {
@@ -463,11 +470,12 @@ const LoraFinetune: FC = observer(() => {
                   return;
                 if (loraParams.loraLoad) {
                   const outputPath = `models/${loraParams.baseModel}-LoRA-${loraParams.loraLoad}`;
-                  MergeLora(commonStore.settings.customPythonPath, true, loraParams.loraAlpha,
+                  MergeLora(commonStore.settings.customPythonPath, !!commonStore.monitorData && commonStore.monitorData.totalVram !== 0, loraParams.loraAlpha,
                     'models/' + loraParams.baseModel, 'lora-models/' + loraParams.loraLoad,
                     outputPath).then(async () => {
                     if (!await FileExists(outputPath)) {
-                      toast(t('Failed to merge model') + ' - ' + await GetPyError(), { type: 'error' });
+                      if (commonStore.platform === 'windows' || commonStore.platform === 'linux')
+                        toast(t('Failed to merge model') + ' - ' + await GetPyError(), { type: 'error' });
                     } else {
                       toast(t('Merge model successfully'), { type: 'success' });
                     }

frontend/src/pages/defaultConfigs.ts

@@ -111,7 +111,7 @@ export const defaultPresets: CompletionPreset[] = [{
 
 export const defaultModelConfigsMac: ModelConfig[] = [
   {
-    name: 'MAC-0.1B-World',
+    name: 'GPU-2G-1B5-World',
     apiParameters: {
       apiPort: 8000,
       maxResponseToken: 4100,
@@ -121,16 +121,15 @@ export const defaultModelConfigsMac: ModelConfig[] = [
       frequencyPenalty: 1
     },
     modelParameters: {
-      modelName: 'RWKV-4-World-0.1B-v1-20230520-ctx4096.pth',
-      device: 'MPS',
-      precision: 'fp32',
+      modelName: 'RWKV-5-World-1B5-v2-20231025-ctx4096.pth',
+      device: 'WebGPU',
+      precision: 'nf4',
       storedLayers: 41,
-      maxStoredLayers: 41,
-      customStrategy: 'mps fp32'
+      maxStoredLayers: 41
     }
   },
   {
-    name: 'MAC-0.4B-World',
+    name: 'GPU-4G-3B-World',
     apiParameters: {
       apiPort: 8000,
       maxResponseToken: 4100,
@@ -140,12 +139,101 @@ export const defaultModelConfigsMac: ModelConfig[] = [
       frequencyPenalty: 1
     },
     modelParameters: {
-      modelName: 'RWKV-4-World-0.4B-v1-20230529-ctx4096.pth',
-      device: 'MPS',
+      modelName: 'RWKV-5-World-3B-v2-20231118-ctx16k.pth',
+      device: 'WebGPU',
+      precision: 'nf4',
+      storedLayers: 41,
+      maxStoredLayers: 41
+    }
+  },
+  {
+    name: 'GPU-4G-3B-CN',
+    apiParameters: {
+      apiPort: 8000,
+      maxResponseToken: 4100,
+      temperature: 1.0,
+      topP: 0.3,
+      presencePenalty: 0,
+      frequencyPenalty: 1
+    },
+    modelParameters: {
+      modelName: 'RWKV-4-World-CHNtuned-3B-v1-20230625-ctx4096.pth',
+      device: 'WebGPU',
+      precision: 'nf4',
+      storedLayers: 41,
+      maxStoredLayers: 41
+    }
+  },
+  {
+    name: 'GPU-7G-7B-World',
+    apiParameters: {
+      apiPort: 8000,
+      maxResponseToken: 4100,
+      temperature: 1.0,
+      topP: 0.3,
+      presencePenalty: 0,
+      frequencyPenalty: 1
+    },
+    modelParameters: {
+      modelName: 'RWKV-4-World-7B-v1-20230626-ctx4096.pth',
+      device: 'WebGPU',
+      precision: 'nf4',
+      storedLayers: 41,
+      maxStoredLayers: 41
+    }
+  },
+  {
+    name: 'GPU-7G-7B-CN',
+    apiParameters: {
+      apiPort: 8000,
+      maxResponseToken: 4100,
+      temperature: 1.0,
+      topP: 0.3,
+      presencePenalty: 0,
+      frequencyPenalty: 1
+    },
+    modelParameters: {
+      modelName: 'RWKV-4-World-CHNtuned-7B-v1-20230709-ctx4096.pth',
+      device: 'WebGPU',
+      precision: 'nf4',
+      storedLayers: 41,
+      maxStoredLayers: 41
+    }
+  },
+  {
+    name: 'CPU-120M-Music',
+    apiParameters: {
+      apiPort: 8000,
+      maxResponseToken: 4100,
+      temperature: 1.0,
+      topP: 0.8,
+      presencePenalty: 0,
+      frequencyPenalty: 1
+    },
+    modelParameters: {
+      modelName: 'RWKV-5-MIDI-120M-v1-20230728-ctx4096.pth',
+      device: 'CPU',
       precision: 'fp32',
       storedLayers: 41,
-      maxStoredLayers: 41,
-      customStrategy: 'mps fp32'
+      maxStoredLayers: 41
+    }
+  },
+  {
+    name: 'CPU-560M-Music',
+    apiParameters: {
+      apiPort: 8000,
+      maxResponseToken: 4100,
+      temperature: 1.0,
+      topP: 0.8,
+      presencePenalty: 0,
+      frequencyPenalty: 1
+    },
+    modelParameters: {
+      modelName: 'RWKV-5-MIDI-560M-v1-20230902-ctx4096.pth',
+      device: 'CPU',
+      precision: 'fp32',
+      storedLayers: 41,
+      maxStoredLayers: 41
     }
   },
   {
@@ -159,26 +247,7 @@ export const defaultModelConfigsMac: ModelConfig[] = [
       frequencyPenalty: 1
     },
     modelParameters: {
-      modelName: 'RWKV-4-World-1.5B-v1-fixed-20230612-ctx4096.pth',
-      device: 'MPS',
-      precision: 'fp32',
-      storedLayers: 41,
-      maxStoredLayers: 41,
-      customStrategy: 'mps fp32'
-    }
-  },
-  {
-    name: 'MAC-1B5-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-1B5-v12-Eng98%-Other2%-20230520-ctx4096.pth',
+      modelName: 'RWKV-5-World-1B5-v2-20231025-ctx4096.pth',
       device: 'MPS',
       precision: 'fp32',
       storedLayers: 41,
@@ -197,26 +266,7 @@ export const defaultModelConfigsMac: ModelConfig[] = [
       frequencyPenalty: 1
     },
     modelParameters: {
-      modelName: 'RWKV-4-World-3B-v1-20230619-ctx4096.pth',
-      device: 'MPS',
-      precision: 'fp32',
-      storedLayers: 41,
-      maxStoredLayers: 41,
-      customStrategy: 'mps fp32'
-    }
-  },
-  {
-    name: 'MAC-3B-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-3B-v12-Eng98%-Other2%-20230520-ctx4096.pth',
+      modelName: 'RWKV-5-World-3B-v2-20231118-ctx16k.pth',
       device: 'MPS',
       precision: 'fp32',
       storedLayers: 41,
@@ -261,190 +311,29 @@ export const defaultModelConfigsMac: ModelConfig[] = [
       maxStoredLayers: 41,
       customStrategy: 'mps fp32'
     }
-  },
-  {
-    name: 'CPU-120M-Music',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.8,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-MIDI-120M-v1-20230714-ctx4096.pth',
-      device: 'CPU',
-      precision: 'fp32',
-      storedLayers: 41,
-      maxStoredLayers: 41
-    }
-  },
-  {
-    name: 'CPU-560M-Music',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.8,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-MIDI-560M-v1-20230717-ctx4096.pth',
-      device: 'CPU',
-      precision: 'fp32',
-      storedLayers: 41,
-      maxStoredLayers: 41
-    }
-  },
-  {
-    name: 'CPU-6G-1B5-World',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-World-1.5B-v1-fixed-20230612-ctx4096.pth',
-      device: 'CPU',
-      precision: 'fp32',
-      storedLayers: 41,
-      maxStoredLayers: 41
-    }
-  },
-  {
-    name: 'CPU-6G-1B5-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-1B5-v12-Eng98%-Other2%-20230520-ctx4096.pth',
-      device: 'CPU',
-      precision: 'fp32',
-      storedLayers: 41,
-      maxStoredLayers: 41
-    }
-  },
-  {
-    name: 'CPU-12G-3B-World',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-World-3B-v1-20230619-ctx4096.pth',
-      device: 'CPU',
-      precision: 'fp32',
-      storedLayers: 41,
-      maxStoredLayers: 41
-    }
-  },
-  {
-    name: 'CPU-12G-3B-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-3B-v12-Eng98%-Other2%-20230520-ctx4096.pth',
-      device: 'CPU',
-      precision: 'fp32',
-      storedLayers: 41,
-      maxStoredLayers: 41
-    }
-  },
-  {
-    name: 'CPU-12G-3B-CN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-World-CHNtuned-3B-v1-20230625-ctx4096.pth',
-      device: 'CPU',
-      precision: 'fp32',
-      storedLayers: 41,
-      maxStoredLayers: 41
-    }
-  },
-  {
-    name: 'CPU-28G-7B-World',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-World-7B-v1-20230626-ctx4096.pth',
-      device: 'CPU',
-      precision: 'fp32',
-      storedLayers: 41,
-      maxStoredLayers: 41
-    }
-  },
-  {
-    name: 'CPU-28G-7B-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-7B-v12-Eng98%-Other2%-20230521-ctx8192.pth',
-      device: 'CPU',
-      precision: 'fp32',
-      storedLayers: 41,
-      maxStoredLayers: 41
-    }
-  },
-  {
-    name: 'CPU-28G-7B-CN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-World-CHNtuned-7B-v1-20230709-ctx4096.pth',
-      device: 'CPU',
-      precision: 'fp32',
-      storedLayers: 41,
-      maxStoredLayers: 41
-    }
   }
 ];
 
 export const defaultModelConfigs: ModelConfig[] = [
+  {
+    name: 'GPU-2G-1B5-World',
+    apiParameters: {
+      apiPort: 8000,
+      maxResponseToken: 4100,
+      temperature: 1.0,
+      topP: 0.3,
+      presencePenalty: 0,
+      frequencyPenalty: 1
+    },
+    modelParameters: {
+      modelName: 'RWKV-5-World-1B5-v2-20231025-ctx4096.pth',
+      device: 'CUDA',
+      precision: 'int8',
+      storedLayers: 41,
+      maxStoredLayers: 41,
+      useCustomCuda: true
+    }
+  },
   {
     name: 'GPU-2G-3B-World',
     apiParameters: {
@@ -456,7 +345,7 @@ export const defaultModelConfigs: ModelConfig[] = [
       frequencyPenalty: 1
     },
     modelParameters: {
-      modelName: 'RWKV-4-World-3B-v1-20230619-ctx4096.pth',
+      modelName: 'RWKV-5-World-3B-v2-20231118-ctx16k.pth',
       device: 'CUDA',
       precision: 'int8',
       storedLayers: 6,
@@ -464,61 +353,6 @@ export const defaultModelConfigs: ModelConfig[] = [
       useCustomCuda: true
     }
   },
-  {
-    name: 'GPU-2G-0.1B-World',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-World-0.1B-v1-20230520-ctx4096.pth',
-      device: 'CUDA',
-      precision: 'fp32', // using fp16 will disable state cache (->)
-      storedLayers: 41,
-      maxStoredLayers: 41
-    }
-  },
-  {
-    name: 'GPU-2G-1B5-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-1B5-v12-Eng98%-Other2%-20230520-ctx4096.pth',
-      device: 'CUDA',
-      precision: 'int8',
-      storedLayers: 4,
-      maxStoredLayers: 41,
-      useCustomCuda: true
-    }
-  },
-  {
-    name: 'GPU-4G-0.4B-World',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-World-0.4B-v1-20230529-ctx4096.pth',
-      device: 'CUDA',
-      precision: 'fp32',
-      storedLayers: 41,
-      maxStoredLayers: 41
-    }
-  },
   {
     name: 'GPU-4G-1B5-World',
     apiParameters: {
@@ -530,27 +364,9 @@ export const defaultModelConfigs: ModelConfig[] = [
       frequencyPenalty: 1
     },
     modelParameters: {
-      modelName: 'RWKV-4-World-1.5B-v1-fixed-20230612-ctx4096.pth',
+      modelName: 'RWKV-5-World-1B5-v2-20231025-ctx4096.pth',
       device: 'CUDA',
-      precision: 'fp32',
-      storedLayers: 8,
-      maxStoredLayers: 41
-    }
-  },
-  {
-    name: 'GPU-4G-1B5-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-1B5-v12-Eng98%-Other2%-20230520-ctx4096.pth',
-      device: 'CUDA',
-      precision: 'int8',
+      precision: 'fp16',
       storedLayers: 41,
       maxStoredLayers: 41,
       useCustomCuda: true
@@ -567,26 +383,7 @@ export const defaultModelConfigs: ModelConfig[] = [
       frequencyPenalty: 1
     },
     modelParameters: {
-      modelName: 'RWKV-4-World-3B-v1-20230619-ctx4096.pth',
-      device: 'CUDA',
-      precision: 'int8',
-      storedLayers: 24,
-      maxStoredLayers: 41,
-      useCustomCuda: true
-    }
-  },
-  {
-    name: 'GPU-4G-3B-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-3B-v12-Eng98%-Other2%-20230520-ctx4096.pth',
+      modelName: 'RWKV-5-World-3B-v2-20231118-ctx16k.pth',
       device: 'CUDA',
       precision: 'int8',
       storedLayers: 24,
@@ -632,25 +429,6 @@ export const defaultModelConfigs: ModelConfig[] = [
       useCustomCuda: true
     }
   },
-  {
-    name: 'GPU-4G-7B-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-7B-v12-Eng98%-Other2%-20230521-ctx8192.pth',
-      device: 'CUDA',
-      precision: 'int8',
-      storedLayers: 8,
-      maxStoredLayers: 41,
-      useCustomCuda: true
-    }
-  },
   {
     name: 'GPU-4G-7B-CN',
     apiParameters: {
@@ -670,25 +448,6 @@ export const defaultModelConfigs: ModelConfig[] = [
       useCustomCuda: true
     }
   },
-  {
-    name: 'GPU-6G-1B5-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-1B5-v12-Eng98%-Other2%-20230520-ctx4096.pth',
-      device: 'CUDA',
-      precision: 'fp16',
-      storedLayers: 41,
-      maxStoredLayers: 41,
-      useCustomCuda: true
-    }
-  },
   {
     name: 'GPU-6G-3B-World',
     apiParameters: {
@@ -700,26 +459,7 @@ export const defaultModelConfigs: ModelConfig[] = [
       frequencyPenalty: 1
     },
     modelParameters: {
-      modelName: 'RWKV-4-World-3B-v1-20230619-ctx4096.pth',
-      device: 'CUDA',
-      precision: 'int8',
-      storedLayers: 41,
-      maxStoredLayers: 41,
-      useCustomCuda: true
-    }
-  },
-  {
-    name: 'GPU-6G-3B-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-3B-v12-Eng98%-Other2%-20230520-ctx4096.pth',
+      modelName: 'RWKV-5-World-3B-v2-20231118-ctx16k.pth',
       device: 'CUDA',
       precision: 'int8',
       storedLayers: 41,
@@ -765,25 +505,6 @@ export const defaultModelConfigs: ModelConfig[] = [
       useCustomCuda: true
     }
   },
-  {
-    name: 'GPU-6G-7B-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-7B-v12-Eng98%-Other2%-20230521-ctx8192.pth',
-      device: 'CUDA',
-      precision: 'int8',
-      storedLayers: 18,
-      maxStoredLayers: 41,
-      useCustomCuda: true
-    }
-  },
   {
     name: 'GPU-6G-7B-CN',
     apiParameters: {
@@ -803,24 +524,6 @@ export const defaultModelConfigs: ModelConfig[] = [
       useCustomCuda: true
     }
   },
-  {
-    name: 'GPU-8G-1B5-World',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-World-1.5B-v1-fixed-20230612-ctx4096.pth',
-      device: 'CUDA',
-      precision: 'fp32',
-      storedLayers: 41,
-      maxStoredLayers: 41
-    }
-  },
   {
     name: 'GPU-8G-3B-World',
     apiParameters: {
@@ -832,26 +535,7 @@ export const defaultModelConfigs: ModelConfig[] = [
       frequencyPenalty: 1
     },
     modelParameters: {
-      modelName: 'RWKV-4-World-3B-v1-20230619-ctx4096.pth',
-      device: 'CUDA',
-      precision: 'fp16',
-      storedLayers: 41,
-      maxStoredLayers: 41,
-      useCustomCuda: true
-    }
-  },
-  {
-    name: 'GPU-8G-3B-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-3B-v12-Eng98%-Other2%-20230520-ctx4096.pth',
+      modelName: 'RWKV-5-World-3B-v2-20231118-ctx16k.pth',
       device: 'CUDA',
       precision: 'fp16',
       storedLayers: 41,
@@ -897,25 +581,6 @@ export const defaultModelConfigs: ModelConfig[] = [
       useCustomCuda: true
     }
   },
-  {
-    name: 'GPU-8G-7B-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-7B-v12-Eng98%-Other2%-20230521-ctx8192.pth',
-      device: 'CUDA',
-      precision: 'int8',
-      storedLayers: 27,
-      maxStoredLayers: 41,
-      useCustomCuda: true
-    }
-  },
   {
     name: 'GPU-8G-7B-CN',
     apiParameters: {
@@ -954,25 +619,6 @@ export const defaultModelConfigs: ModelConfig[] = [
       useCustomCuda: true
     }
   },
-  {
-    name: 'GPU-10G-7B-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-7B-v12-Eng98%-Other2%-20230521-ctx8192.pth',
-      device: 'CUDA',
-      precision: 'int8',
-      storedLayers: 41,
-      maxStoredLayers: 41,
-      useCustomCuda: true
-    }
-  },
   {
     name: 'GPU-10G-7B-CN',
     apiParameters: {
@@ -992,25 +638,6 @@ export const defaultModelConfigs: ModelConfig[] = [
       useCustomCuda: true
     }
   },
-  {
-    name: 'GPU-12G-14B-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-14B-v12-Eng98%-Other2%-20230523-ctx8192.pth',
-      device: 'CUDA',
-      precision: 'int8',
-      storedLayers: 24,
-      maxStoredLayers: 41,
-      useCustomCuda: true
-    }
-  },
   {
     name: 'GPU-16G-7B-World',
     apiParameters: {
@@ -1030,25 +657,6 @@ export const defaultModelConfigs: ModelConfig[] = [
       useCustomCuda: true
     }
   },
-  {
-    name: 'GPU-16G-7B-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-7B-v12-Eng98%-Other2%-20230521-ctx8192.pth',
-      device: 'CUDA',
-      precision: 'fp16',
-      storedLayers: 41,
-      maxStoredLayers: 41,
-      useCustomCuda: true
-    }
-  },
   {
     name: 'GPU-16G-7B-CN',
     apiParameters: {
@@ -1068,63 +676,6 @@ export const defaultModelConfigs: ModelConfig[] = [
       useCustomCuda: true
     }
   },
-  {
-    name: 'GPU-16G-14B-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-14B-v12-Eng98%-Other2%-20230523-ctx8192.pth',
-      device: 'CUDA',
-      precision: 'int8',
-      storedLayers: 37,
-      maxStoredLayers: 41,
-      useCustomCuda: true
-    }
-  },
-  {
-    name: 'GPU-18G-14B-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-14B-v12-Eng98%-Other2%-20230523-ctx8192.pth',
-      device: 'CUDA',
-      precision: 'int8',
-      storedLayers: 41,
-      maxStoredLayers: 41,
-      useCustomCuda: true
-    }
-  },
-  {
-    name: 'GPU-32G-14B-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-14B-v12-Eng98%-Other2%-20230523-ctx8192.pth',
-      device: 'CUDA',
-      precision: 'fp16',
-      storedLayers: 41,
-      maxStoredLayers: 41,
-      useCustomCuda: true
-    }
-  },
   {
     name: 'CPU-120M-Music',
     apiParameters: {
@@ -1136,7 +687,7 @@ export const defaultModelConfigs: ModelConfig[] = [
       frequencyPenalty: 1
     },
     modelParameters: {
-      modelName: 'RWKV-4-MIDI-120M-v1-20230714-ctx4096.pth',
+      modelName: 'RWKV-5-MIDI-120M-v1-20230728-ctx4096.pth',
       device: 'CPU',
       precision: 'fp32',
       storedLayers: 41,
@@ -1154,7 +705,7 @@ export const defaultModelConfigs: ModelConfig[] = [
       frequencyPenalty: 1
     },
     modelParameters: {
-      modelName: 'RWKV-4-MIDI-560M-v1-20230717-ctx4096.pth',
+      modelName: 'RWKV-5-MIDI-560M-v1-20230902-ctx4096.pth',
       device: 'CPU',
       precision: 'fp32',
       storedLayers: 41,
@@ -1162,7 +713,7 @@ export const defaultModelConfigs: ModelConfig[] = [
     }
   },
   {
-    name: 'CPU-6G-1B5-World',
+    name: 'AnyGPU-2G-1B5-World',
     apiParameters: {
       apiPort: 8000,
       maxResponseToken: 4100,
@@ -1172,15 +723,15 @@ export const defaultModelConfigs: ModelConfig[] = [
       frequencyPenalty: 1
     },
     modelParameters: {
-      modelName: 'RWKV-4-World-1.5B-v1-fixed-20230612-ctx4096.pth',
-      device: 'CPU',
-      precision: 'fp32',
+      modelName: 'RWKV-5-World-1B5-v2-20231025-ctx4096.pth',
+      device: 'WebGPU',
+      precision: 'nf4',
       storedLayers: 41,
       maxStoredLayers: 41
     }
   },
   {
-    name: 'CPU-6G-1B5-EN',
+    name: 'AnyGPU-4G-3B-World',
     apiParameters: {
       apiPort: 8000,
       maxResponseToken: 4100,
@@ -1190,51 +741,15 @@ export const defaultModelConfigs: ModelConfig[] = [
       frequencyPenalty: 1
     },
     modelParameters: {
-      modelName: 'RWKV-4-Raven-1B5-v12-Eng98%-Other2%-20230520-ctx4096.pth',
-      device: 'CPU',
-      precision: 'fp32',
+      modelName: 'RWKV-5-World-3B-v2-20231118-ctx16k.pth',
+      device: 'WebGPU',
+      precision: 'nf4',
       storedLayers: 41,
       maxStoredLayers: 41
     }
   },
   {
-    name: 'CPU-12G-3B-World',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-World-3B-v1-20230619-ctx4096.pth',
-      device: 'CPU',
-      precision: 'fp32',
-      storedLayers: 41,
-      maxStoredLayers: 41
-    }
-  },
-  {
-    name: 'CPU-12G-3B-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-3B-v12-Eng98%-Other2%-20230520-ctx4096.pth',
-      device: 'CPU',
-      precision: 'fp32',
-      storedLayers: 41,
-      maxStoredLayers: 41
-    }
-  },
-  {
-    name: 'CPU-12G-3B-CN',
+    name: 'AnyGPU-4G-3B-CN',
     apiParameters: {
       apiPort: 8000,
       maxResponseToken: 4100,
@@ -1245,14 +760,14 @@ export const defaultModelConfigs: ModelConfig[] = [
     },
     modelParameters: {
       modelName: 'RWKV-4-World-CHNtuned-3B-v1-20230625-ctx4096.pth',
-      device: 'CPU',
-      precision: 'fp32',
+      device: 'WebGPU',
+      precision: 'nf4',
       storedLayers: 41,
       maxStoredLayers: 41
     }
   },
   {
-    name: 'CPU-28G-7B-World',
+    name: 'AnyGPU-7G-7B-World',
     apiParameters: {
       apiPort: 8000,
       maxResponseToken: 4100,
@@ -1263,32 +778,14 @@ export const defaultModelConfigs: ModelConfig[] = [
     },
     modelParameters: {
       modelName: 'RWKV-4-World-7B-v1-20230626-ctx4096.pth',
-      device: 'CPU',
-      precision: 'fp32',
+      device: 'WebGPU',
+      precision: 'nf4',
       storedLayers: 41,
       maxStoredLayers: 41
     }
   },
   {
-    name: 'CPU-28G-7B-EN',
-    apiParameters: {
-      apiPort: 8000,
-      maxResponseToken: 4100,
-      temperature: 1.0,
-      topP: 0.3,
-      presencePenalty: 0,
-      frequencyPenalty: 1
-    },
-    modelParameters: {
-      modelName: 'RWKV-4-Raven-7B-v12-Eng98%-Other2%-20230521-ctx8192.pth',
-      device: 'CPU',
-      precision: 'fp32',
-      storedLayers: 41,
-      maxStoredLayers: 41
-    }
-  },
-  {
-    name: 'CPU-28G-7B-CN',
+    name: 'AnyGPU-7G-7B-CN',
     apiParameters: {
       apiPort: 8000,
       maxResponseToken: 4100,
@@ -1299,8 +796,8 @@ export const defaultModelConfigs: ModelConfig[] = [
     },
     modelParameters: {
       modelName: 'RWKV-4-World-CHNtuned-7B-v1-20230709-ctx4096.pth',
-      device: 'CPU',
-      precision: 'fp32',
+      device: 'WebGPU',
+      precision: 'nf4',
       storedLayers: 41,
       maxStoredLayers: 41
     }

frontend/src/startup.ts

@@ -1,5 +1,5 @@
 import commonStore, { MonitorData, Platform } from './stores/commonStore';
-import { GetPlatform, ListDirFiles, ReadJson } from '../wailsjs/go/backend_golang/App';
+import { FileExists, GetPlatform, ListDirFiles, ReadJson } from '../wailsjs/go/backend_golang/App';
 import { Cache, checkUpdate, downloadProgramFiles, LocalConfig, refreshLocalModels, refreshModels } from './utils';
 import { getStatus } from './apis';
 import { EventsOn, WindowSetTitle } from '../wailsjs/runtime';
@@ -7,6 +7,8 @@ import manifest from '../../manifest.json';
 import { defaultModelConfigs, defaultModelConfigsMac } from './pages/defaultConfigs';
 import { t } from 'i18next';
 import { Preset } from './types/presets';
+import { toast } from 'react-toastify';
+import { MidiMessage, MidiPort } from './types/composition';
 
 export async function startup() {
   initPresets();
@@ -26,6 +28,7 @@ export async function startup() {
     initLocalModelsNotify();
     initLoraModels();
     initHardwareMonitor();
+    initMidi();
   }
 
   await initConfig();
@@ -134,3 +137,22 @@ async function initHardwareMonitor() {
     }
   });
 }
+
+async function initMidi() {
+  EventsOn('midiError', (data: string) => {
+    if (commonStore.platform === 'windows')
+      toast('MIDI Error: ' + data, { type: 'error' });
+  });
+  EventsOn('midiPorts', (data: MidiPort[]) => {
+    commonStore.setMidiPorts(data);
+  });
+  EventsOn('midiMessage', async (data: MidiMessage) => {
+    await (await import('./pages/AudiotrackManager/AudiotrackEditor')).midiMessageHandler(data);
+  });
+  if (await FileExists('assets/sound-font/accordion/instrument.json')) {
+    commonStore.setCompositionParams({
+      ...commonStore.compositionParams,
+      useLocalSoundFont: true
+    });
+  }
+}

frontend/src/stores/commonStore.ts

@@ -7,9 +7,16 @@ import { defaultCompositionPrompt, defaultModelConfigs, defaultModelConfigsMac }
 import { ChartData } from 'chart.js';
 import { Preset } from '../types/presets';
 import { AboutContent } from '../types/about';
-import { Attachment, Conversation } from '../types/chat';
+import { Attachment, ChatParams, Conversation } from '../types/chat';
 import { CompletionPreset } from '../types/completion';
-import { CompositionParams } from '../types/composition';
+import {
+  CompositionParams,
+  InstrumentType,
+  MidiMessage,
+  MidiPort,
+  Track,
+  tracksMinimalTotalTime
+} from '../types/composition';
 import { ModelConfig } from '../types/configs';
 import { DownloadStatus } from '../types/downloads';
 import { IntroductionContent } from '../types/home';
@@ -51,6 +58,7 @@ class CommonStore {
   monitorData: MonitorData | null = null;
   depComplete: boolean = false;
   platform: Platform = 'windows';
+  lastModelName: string = '';
   // presets manager
   editingPreset: Preset | null = null;
   presets: Preset[] = [];
@@ -62,8 +70,18 @@ class CommonStore {
   conversationOrder: string[] = [];
   activePreset: Preset | null = null;
   attachmentUploading: boolean = false;
-  attachments: { [uuid: string]: Attachment[] } = {};
+  attachments: {
+    [uuid: string]: Attachment[]
+  } = {};
   currentTempAttachment: Attachment | null = null;
+  chatParams: ChatParams = {
+    maxResponseToken: 1000,
+    temperature: 1,
+    topP: 0.3,
+    presencePenalty: 0,
+    frequencyPenalty: 1
+  };
+  sidePanelCollapsed: boolean | 'auto' = 'auto';
   // completion
   completionPreset: CompletionPreset | null = null;
   completionGenerating: boolean = false;
@@ -76,16 +94,32 @@ class CommonStore {
     topP: 0.8,
     autoPlay: true,
     useLocalSoundFont: false,
+    externalPlay: false,
     midi: null,
     ns: null
   };
   compositionGenerating: boolean = false;
   compositionSubmittedPrompt: string = defaultCompositionPrompt;
+  // composition midi device
+  midiPorts: MidiPort[] = [];
+  activeMidiDeviceIndex: number = -1;
+  instrumentType: InstrumentType = InstrumentType.Piano;
+  // composition tracks
+  tracks: Track[] = [];
+  trackScale: number = 1;
+  trackTotalTime: number = tracksMinimalTotalTime;
+  trackCurrentTime: number = 0;
+  trackPlayStartTime: number = 0;
+  playingTrackId: string = '';
+  recordingTrackId: string = '';
+  recordingContent: string = ''; // used to improve performance of midiMessageHandler, and I'm too lazy to maintain an ID dictionary for this (although that would be better for realtime effects)
+  recordingRawContent: MidiMessage[] = [];
   // configs
   currentModelConfigIndex: number = 0;
   modelConfigs: ModelConfig[] = [];
   modelParamsCollapsed: boolean = true;
   // models
+  activeModelListTags: string[] = [];
   modelSourceManifestList: string = 'https://cdn.jsdelivr.net/gh/josstorer/RWKV-Runner@master/manifest.json;';
   modelSourceList: ModelSourceItem[] = [];
   // downloads
@@ -133,6 +167,7 @@ class CommonStore {
     autoUpdatesCheck: true,
     giteeUpdatesSource: getUserLanguage() === 'zh',
     cnMirror: getUserLanguage() === 'zh',
+    useHfMirror: false,
     host: '127.0.0.1',
     dpiScaling: 100,
     customModelsPath: './models',
@@ -140,7 +175,8 @@ class CommonStore {
     apiUrl: '',
     apiKey: '',
     apiChatModelName: 'rwkv',
-    apiCompletionModelName: 'rwkv'
+    apiCompletionModelName: 'rwkv',
+    coreApiUrl: ''
   };
   // about
   about: AboutContent = manifest.about;
@@ -232,6 +268,10 @@ class CommonStore {
     this.about = value;
   };
 
+  setLastModelName(value: string) {
+    this.lastModelName = value;
+  }
+
   setDepComplete = (value: boolean, inSaveCache: boolean = true) => {
     this.depComplete = value;
     if (inSaveCache)
@@ -292,7 +332,7 @@ class CommonStore {
       savePresets();
   }
 
-  setActivePreset(value: Preset) {
+  setActivePreset(value: Preset | null) {
     this.activePreset = value;
   }
 
@@ -344,7 +384,9 @@ class CommonStore {
     this.attachmentUploading = value;
   }
 
-  setAttachments(value: { [uuid: string]: Attachment[] }) {
+  setAttachments(value: {
+    [uuid: string]: Attachment[]
+  }) {
     this.attachments = value;
   }
 
@@ -358,6 +400,66 @@ class CommonStore {
   setCurrentTempAttachment(value: Attachment | null) {
     this.currentTempAttachment = value;
   }
+
+  setChatParams(value: Partial<ChatParams>) {
+    this.chatParams = { ...this.chatParams, ...value };
+  }
+
+  setSidePanelCollapsed(value: boolean | 'auto') {
+    this.sidePanelCollapsed = value;
+  }
+
+  setTracks(value: Track[]) {
+    this.tracks = value;
+  }
+
+  setTrackScale(value: number) {
+    this.trackScale = value;
+  }
+
+  setTrackTotalTime(value: number) {
+    this.trackTotalTime = value;
+  }
+
+  setTrackCurrentTime(value: number) {
+    this.trackCurrentTime = value;
+  }
+
+  setTrackPlayStartTime(value: number) {
+    this.trackPlayStartTime = value;
+  }
+
+  setMidiPorts(value: MidiPort[]) {
+    this.midiPorts = value;
+  }
+
+  setInstrumentType(value: InstrumentType) {
+    this.instrumentType = value;
+  }
+
+  setRecordingTrackId(value: string) {
+    this.recordingTrackId = value;
+  }
+
+  setActiveMidiDeviceIndex(value: number) {
+    this.activeMidiDeviceIndex = value;
+  }
+
+  setRecordingContent(value: string) {
+    this.recordingContent = value;
+  }
+
+  setRecordingRawContent(value: MidiMessage[]) {
+    this.recordingRawContent = value;
+  }
+
+  setPlayingTrackId(value: string) {
+    this.playingTrackId = value;
+  }
+
+  setActiveModelListTags(value: string[]) {
+    this.activeModelListTags = value;
+  }
 }
 
 export default new CommonStore();

frontend/src/types/chat.ts

@@ -1,3 +1,5 @@
+import { ApiParameters } from './configs';
+
 export const userName = 'M E';
 export const botName = 'A I';
 export const welcomeUuid = 'welcome';
@@ -31,4 +33,5 @@ export type Attachment = {
   name: string;
   size: number;
   content: string;
-}
+}
+export type ChatParams = Omit<ApiParameters, 'apiPort'>

frontend/src/types/composition.ts

@@ -1,5 +1,7 @@
 import { NoteSequence } from '@magenta/music/esm/protobuf';
 
+export const tracksMinimalTotalTime = 5000;
+
 export type CompositionParams = {
   prompt: string,
   maxResponseToken: number,
@@ -7,6 +9,78 @@ export type CompositionParams = {
   topP: number,
   autoPlay: boolean,
   useLocalSoundFont: boolean,
+  externalPlay: boolean,
   midi: ArrayBuffer | null,
   ns: NoteSequence | null
-}
+}
+export type Track = {
+  id: string;
+  mainInstrument: string;
+  content: string;
+  rawContent: MidiMessage[];
+  offsetTime: number;
+  contentTime: number;
+};
+export type MidiPort = {
+  name: string;
+}
+
+export type MessageType = 'NoteOff' | 'NoteOn' | 'ElapsedTime' | 'ControlChange';
+
+export type MidiMessage = {
+  messageType: MessageType;
+  channel: number;
+  note: number;
+  velocity: number;
+  control: number;
+  value: number;
+  instrument: InstrumentType;
+}
+
+export enum InstrumentType {
+  Piano,
+  Percussion,
+  Drum,
+  Tuba,
+  Marimba,
+  Bass,
+  Guitar,
+  Violin,
+  Trumpet,
+  Sax,
+  Flute,
+  Lead,
+  Pad,
+}
+
+export const InstrumentTypeNameMap = [
+  'Piano',
+  'Percussion',
+  'Drum',
+  'Tuba',
+  'Marimba',
+  'Bass',
+  'Guitar',
+  'Violin',
+  'Trumpet',
+  'Sax',
+  'Flute',
+  'Lead',
+  'Pad'
+];
+
+export const InstrumentTypeTokenMap = [
+  'pi',
+  'p',
+  'd',
+  't',
+  'm',
+  'b',
+  'g',
+  'v',
+  'tr',
+  's',
+  'f',
+  'l',
+  'pa'
+];

frontend/src/types/configs.ts

@@ -6,8 +6,8 @@ export type ApiParameters = {
   presencePenalty: number;
   frequencyPenalty: number;
 }
-export type Device = 'CPU' | 'CUDA' | 'CUDA-Beta' | 'WebGPU' | 'MPS' | 'Custom';
-export type Precision = 'fp16' | 'int8' | 'fp32';
+export type Device = 'CPU' | 'CPU (rwkv.cpp)' | 'CUDA' | 'CUDA-Beta' | 'WebGPU' | 'WebGPU (Python)' | 'MPS' | 'Custom';
+export type Precision = 'fp16' | 'int8' | 'fp32' | 'nf4' | 'Q5_1';
 export type ModelParameters = {
   // different models can not have the same name
   modelName: string;

frontend/src/types/models.ts

@@ -10,5 +10,6 @@ export type ModelSourceItem = {
   isLocal?: boolean;
   localSize?: number;
   lastUpdatedMs?: number;
+  tags?: string[];
   hide?: boolean;
 };

frontend/src/types/settings.ts

@@ -10,6 +10,7 @@ export type SettingsType = {
   autoUpdatesCheck: boolean
   giteeUpdatesSource: boolean
   cnMirror: boolean
+  useHfMirror: boolean
   host: string
   dpiScaling: number
   customModelsPath: string
@@ -18,4 +19,5 @@ export type SettingsType = {
   apiKey: string
   apiChatModelName: string
   apiCompletionModelName: string
+  coreApiUrl: string
 }

frontend/src/utils/convert-model.ts

@@ -0,0 +1,118 @@
+import { toast } from 'react-toastify';
+import commonStore from '../stores/commonStore';
+import { t } from 'i18next';
+import {
+  ConvertGGML,
+  ConvertModel,
+  ConvertSafetensors,
+  ConvertSafetensorsWithPython,
+  FileExists,
+  GetPyError
+} from '../../wailsjs/go/backend_golang/App';
+import { WindowShow } from '../../wailsjs/runtime';
+import { ModelConfig, Precision } from '../types/configs';
+import { checkDependencies, getStrategy } from './index';
+import { NavigateFunction } from 'react-router';
+
+export const convertModel = async (selectedConfig: ModelConfig, navigate: NavigateFunction) => {
+  if (commonStore.platform === 'darwin') {
+    toast(t('MacOS is not yet supported for performing this operation, please do it manually.') + ' (backend-python/convert_model.py)', { type: 'info' });
+    return;
+  } else if (commonStore.platform === 'linux') {
+    toast(t('Linux is not yet supported for performing this operation, please do it manually.') + ' (backend-python/convert_model.py)', { type: 'info' });
+    return;
+  }
+
+  const ok = await checkDependencies(navigate);
+  if (!ok)
+    return;
+
+  const modelPath = `${commonStore.settings.customModelsPath}/${selectedConfig.modelParameters.modelName}`;
+  if (await FileExists(modelPath)) {
+    const strategy = getStrategy(selectedConfig);
+    const newModelPath = modelPath + '-' + strategy.replace(/[:> *+]/g, '-');
+    toast(t('Start Converting'), { autoClose: 2000, type: 'info' });
+    ConvertModel(commonStore.settings.customPythonPath, modelPath, strategy, newModelPath).then(async () => {
+      if (!await FileExists(newModelPath + '.pth')) {
+        toast(t('Convert Failed') + ' - ' + await GetPyError(), { type: 'error' });
+      } else {
+        toast(`${t('Convert Success')} - ${newModelPath}`, { type: 'success' });
+      }
+    }).catch(e => {
+      const errMsg = e.message || e;
+      if (errMsg.includes('path contains space'))
+        toast(`${t('Convert Failed')} - ${t('File Path Cannot Contain Space')}`, { type: 'error' });
+      else
+        toast(`${t('Convert Failed')} - ${e.message || e}`, { type: 'error' });
+    });
+    setTimeout(WindowShow, 1000);
+  } else {
+    toast(`${t('Model Not Found')} - ${modelPath}`, { type: 'error' });
+  }
+};
+
+
+export const convertToSt = async (selectedConfig: ModelConfig, navigate: NavigateFunction) => {
+  const webgpuPython = selectedConfig.modelParameters.device === 'WebGPU (Python)';
+  if (webgpuPython) {
+    const ok = await checkDependencies(navigate);
+    if (!ok)
+      return;
+  }
+
+  const modelPath = `${commonStore.settings.customModelsPath}/${selectedConfig.modelParameters.modelName}`;
+  if (await FileExists(modelPath)) {
+    toast(t('Start Converting'), { autoClose: 2000, type: 'info' });
+    const newModelPath = modelPath.replace(/\.pth$/, '.st');
+    const convert = webgpuPython ?
+      (input: string, output: string) => ConvertSafetensorsWithPython(commonStore.settings.customPythonPath, input, output)
+      : ConvertSafetensors;
+    convert(modelPath, newModelPath).then(async () => {
+      if (!await FileExists(newModelPath)) {
+        if (commonStore.platform === 'windows' || commonStore.platform === 'linux')
+          toast(t('Convert Failed') + ' - ' + await GetPyError(), { type: 'error' });
+      } else {
+        toast(`${t('Convert Success')} - ${newModelPath}`, { type: 'success' });
+      }
+    }).catch(e => {
+      const errMsg = e.message || e;
+      if (errMsg.includes('path contains space'))
+        toast(`${t('Convert Failed')} - ${t('File Path Cannot Contain Space')}`, { type: 'error' });
+      else
+        toast(`${t('Convert Failed')} - ${e.message || e}`, { type: 'error' });
+    });
+    setTimeout(WindowShow, 1000);
+  } else {
+    toast(`${t('Model Not Found')} - ${modelPath}`, { type: 'error' });
+  }
+};
+
+export const convertToGGML = async (selectedConfig: ModelConfig, navigate: NavigateFunction) => {
+  const ok = await checkDependencies(navigate);
+  if (!ok)
+    return;
+
+  const modelPath = `${commonStore.settings.customModelsPath}/${selectedConfig.modelParameters.modelName}`;
+  if (await FileExists(modelPath)) {
+    toast(t('Start Converting'), { autoClose: 2000, type: 'info' });
+    const precision: Precision = selectedConfig.modelParameters.precision === 'Q5_1' ? 'Q5_1' : 'fp16';
+    const newModelPath = modelPath.replace(/\.pth$/, `-${precision}.bin`);
+    ConvertGGML(commonStore.settings.customPythonPath, modelPath, newModelPath, precision === 'Q5_1').then(async () => {
+      if (!await FileExists(newModelPath)) {
+        if (commonStore.platform === 'windows' || commonStore.platform === 'linux')
+          toast(t('Convert Failed') + ' - ' + await GetPyError(), { type: 'error' });
+      } else {
+        toast(`${t('Convert Success')} - ${newModelPath}`, { type: 'success' });
+      }
+    }).catch(e => {
+      const errMsg = e.message || e;
+      if (errMsg.includes('path contains space'))
+        toast(`${t('Convert Failed')} - ${t('File Path Cannot Contain Space')}`, { type: 'error' });
+      else
+        toast(`${t('Convert Failed')} - ${e.message || e}`, { type: 'error' });
+    });
+    setTimeout(WindowShow, 1000);
+  } else {
+    toast(`${t('Model Not Found')} - ${modelPath}`, { type: 'error' });
+  }
+};

frontend/src/utils/index.tsx

@@ -22,6 +22,11 @@ import { DownloadStatus } from '../types/downloads';
 import { ModelSourceItem } from '../types/models';
 import { Language, Languages, SettingsType } from '../types/settings';
 import { DataProcessParameters, LoraFinetuneParameters } from '../types/train';
+import { InstrumentTypeNameMap, MidiMessage, tracksMinimalTotalTime } from '../types/composition';
+import logo from '../assets/images/logo.png';
+import { Preset } from '../types/presets';
+import { botName, Conversation, MessageType, userName } from '../types/chat';
+import { v4 as uuid } from 'uuid';
 
 export type Cache = {
   version: string
@@ -39,16 +44,18 @@ export type LocalConfig = {
 }
 
 export async function refreshBuiltInModels(readCache: boolean = false) {
-  let cache: { models: ModelSourceItem[] } = { models: [] };
+  let cache: {
+    models: ModelSourceItem[]
+  } = { models: [] };
   if (readCache)
     await ReadJson('cache.json').then((cacheData: Cache) => {
       if (cacheData.models)
         cache.models = cacheData.models;
-      else cache.models = manifest.models;
+      else cache.models = manifest.models.slice();
     }).catch(() => {
-      cache.models = manifest.models;
+      cache.models = manifest.models.slice();
     });
-  else cache.models = manifest.models;
+  else cache.models = manifest.models.slice();
 
   commonStore.setModelSourceList(cache.models);
   await saveCache().catch(() => {
@@ -56,7 +63,7 @@ export async function refreshBuiltInModels(readCache: boolean = false) {
   return cache;
 }
 
-const modelSuffix = ['.pth', '.st', '.safetensors'];
+const modelSuffix = ['.pth', '.st', '.safetensors', '.bin'];
 
 export async function refreshLocalModels(cache: {
   models: ModelSourceItem[]
@@ -72,7 +79,8 @@ export async function refreshLocalModels(cache: {
           size: d.size,
           lastUpdated: d.modTime,
           isComplete: true,
-          isLocal: true
+          isLocal: true,
+          tags: ['Local']
         }] as ModelSourceItem[];
       return [];
     }));
@@ -82,12 +90,15 @@ export async function refreshLocalModels(cache: {
   for (let i = 0; i < cache.models.length; i++) {
     if (!cache.models[i].lastUpdatedMs)
       cache.models[i].lastUpdatedMs = Date.parse(cache.models[i].lastUpdated);
+    if (!cache.models[i].tags)
+      cache.models[i].tags = ['Other'];
 
     for (let j = i + 1; j < cache.models.length; j++) {
       if (!cache.models[j].lastUpdatedMs)
         cache.models[j].lastUpdatedMs = Date.parse(cache.models[j].lastUpdated);
 
       if (cache.models[i].name === cache.models[j].name) {
+        const tags = Array.from(new Set([...cache.models[i].tags as string[], ...cache.models[j].tags as string[]]));
         if (cache.models[i].size <= cache.models[j].size) { // j is local file
           if (cache.models[i].lastUpdatedMs! < cache.models[j].lastUpdatedMs!) {
             cache.models[i] = Object.assign({}, cache.models[i], cache.models[j]);
@@ -97,6 +108,7 @@ export async function refreshLocalModels(cache: {
         } // else is not complete local file
         cache.models[i].isLocal = true;
         cache.models[i].localSize = cache.models[j].size;
+        cache.models[i].tags = tags;
         cache.models.splice(j, 1);
         j--;
       }
@@ -117,7 +129,7 @@ function initLastUnfinishedModelDownloads() {
         {
           name: item.name,
           path: `${commonStore.settings.customModelsPath}/${item.name}`,
-          url: item.downloadUrl!,
+          url: getHfDownloadUrl(item.downloadUrl!),
           transferred: item.localSize!,
           size: item.size,
           speed: 0,
@@ -131,7 +143,9 @@ function initLastUnfinishedModelDownloads() {
   commonStore.setLastUnfinishedModelDownloads(list);
 }
 
-export async function refreshRemoteModels(cache: { models: ModelSourceItem[] }) {
+export async function refreshRemoteModels(cache: {
+  models: ModelSourceItem[]
+}) {
   const manifestUrls = commonStore.modelSourceManifestList.split(/[,，;；\n]/);
   const requests = manifestUrls.filter(url => url.endsWith('.json')).map(
     url => fetch(url, { cache: 'no-cache' }).then(r => r.json()));
@@ -178,14 +192,15 @@ export const getStrategy = (modelConfig: ModelConfig | undefined = undefined) =>
       strategy += params.precision === 'int8' ? 'fp32i8' : 'fp32';
       break;
     case 'WebGPU':
-      strategy += params.precision === 'int8' ? 'fp16i8' : 'fp16';
+    case 'WebGPU (Python)':
+      strategy += params.precision === 'nf4' ? 'fp16i4' : params.precision === 'int8' ? 'fp16i8' : 'fp16';
       break;
     case 'CUDA':
     case 'CUDA-Beta':
       if (avoidOverflow)
         strategy = params.useCustomCuda ? 'cuda fp16 *1 -> ' : 'cuda fp32 *1 -> ';
       strategy += 'cuda ';
-      strategy += params.precision === 'fp16' ? 'fp16' : params.precision === 'int8' ? 'fp16i8' : 'fp32';
+      strategy += params.precision === 'int8' ? 'fp16i8' : params.precision === 'fp32' ? 'fp32' : 'fp16';
       if (params.storedLayers < params.maxStoredLayers)
         strategy += ` *${params.storedLayers}+`;
       break;
@@ -193,7 +208,7 @@ export const getStrategy = (modelConfig: ModelConfig | undefined = undefined) =>
       if (avoidOverflow)
         strategy = 'mps fp32 *1 -> ';
       strategy += 'mps ';
-      strategy += params.precision === 'fp16' ? 'fp16' : params.precision === 'int8' ? 'fp16i8' : 'fp32';
+      strategy += params.precision === 'int8' ? 'fp32i8' : 'fp32';
       break;
     case 'Custom':
       strategy = params.customStrategy || '';
@@ -289,18 +304,27 @@ export function bytesToReadable(size: number) {
   else return bytesToGb(size) + ' GB';
 }
 
-export function getServerRoot(defaultLocalPort: number) {
+export function getServerRoot(defaultLocalPort: number, isCore: boolean = false) {
+  const coreCustomApiUrl = commonStore.settings.coreApiUrl.trim().replace(/\/$/, '');
+  if (isCore && coreCustomApiUrl)
+    return coreCustomApiUrl;
+
+  const defaultRoot = `http://127.0.0.1:${defaultLocalPort}`;
+  if (commonStore.status.status !== ModelStatus.Offline)
+    return defaultRoot;
   const customApiUrl = commonStore.settings.apiUrl.trim().replace(/\/$/, '');
   if (customApiUrl)
     return customApiUrl;
   if (commonStore.platform === 'web')
     return '';
-  return `http://127.0.0.1:${defaultLocalPort}`;
+  return defaultRoot;
 }
 
 export function absPathAsset(path: string) {
   if (commonStore.platform === 'web')
     return path;
+  if (path === logo)
+    return path;
   if ((path.length > 0 && path[0] === '/') ||
     (path.length > 1 && path[1] === ':')) {
     return '=>' + path;
@@ -414,7 +438,10 @@ export const checkDependencies = async (navigate: NavigateFunction) => {
           toastWithButton(`${t('Downloading')} Python`, t('Check'), () => {
             navigate({ pathname: '/downloads' });
           }, { autoClose: 3000 });
-          AddToDownloadList('python-3.10.11-embed-amd64.zip', 'https://www.python.org/ftp/python/3.10.11/python-3.10.11-embed-amd64.zip');
+          AddToDownloadList('python-3.10.11-embed-amd64.zip',
+            !commonStore.settings.cnMirror
+              ? 'https://www.python.org/ftp/python/3.10.11/python-3.10.11-embed-amd64.zip'
+              : 'https://mirrors.huaweicloud.com/python/3.10.11/python-3.10.11-embed-amd64.zip');
         });
       } else if (depErrorMsg.includes('DepCheck Error')) {
         if (depErrorMsg.includes('vc_redist') || depErrorMsg.includes('DLL load failed while importing')) {
@@ -471,6 +498,107 @@ export function toastWithButton(text: string, buttonText: string, onClickButton:
   return id;
 }
 
+export function getHfDownloadUrl(url: string) {
+  if (commonStore.settings.useHfMirror && url.includes('huggingface.co') && url.includes('resolve'))
+    return url.replace('huggingface.co', 'hf-mirror.com');
+  return url;
+}
+
+export function refreshTracksTotalTime() {
+  if (commonStore.tracks.length === 0) {
+    commonStore.setTrackTotalTime(tracksMinimalTotalTime);
+    commonStore.setTrackCurrentTime(0);
+    commonStore.setTrackPlayStartTime(0);
+    return;
+  }
+  const endTimes = commonStore.tracks.map(t => t.offsetTime + t.contentTime);
+  const totalTime = Math.max(...endTimes) + tracksMinimalTotalTime;
+  if (commonStore.trackPlayStartTime > totalTime)
+    commonStore.setTrackPlayStartTime(totalTime);
+  commonStore.setTrackTotalTime(totalTime);
+}
+
+export function getMidiRawContentTime(rawContent: MidiMessage[]) {
+  return rawContent.reduce((sum, current) =>
+      sum + (current.messageType === 'ElapsedTime' ? current.value : 0)
+    , 0);
+}
+
+export function getMidiRawContentMainInstrument(rawContent: MidiMessage[]) {
+  const sortedInstrumentFrequency = Object.entries(rawContent
+  .filter(c => c.messageType === 'NoteOn')
+  .map(c => c.instrument)
+  .reduce((frequencyCount, current) => (frequencyCount[current] = (frequencyCount[current] || 0) + 1, frequencyCount)
+    , {} as {
+      [key: string]: number
+    }))
+  .sort((a, b) => b[1] - a[1]);
+  let mainInstrument: string = '';
+  if (sortedInstrumentFrequency.length > 0)
+    mainInstrument = InstrumentTypeNameMap[Number(sortedInstrumentFrequency[0][0])];
+  return mainInstrument;
+}
+
+export function flushMidiRecordingContent() {
+  const recordingTrackIndex = commonStore.tracks.findIndex(t => t.id === commonStore.recordingTrackId);
+  if (recordingTrackIndex >= 0) {
+    const recordingTrack = commonStore.tracks[recordingTrackIndex];
+    const tracks = commonStore.tracks.slice();
+
+    tracks[recordingTrackIndex] = {
+      ...recordingTrack,
+      content: commonStore.recordingContent,
+      rawContent: commonStore.recordingRawContent,
+      contentTime: getMidiRawContentTime(commonStore.recordingRawContent),
+      mainInstrument: getMidiRawContentMainInstrument(commonStore.recordingRawContent)
+    };
+    commonStore.setTracks(tracks);
+    refreshTracksTotalTime();
+  }
+  commonStore.setRecordingContent('');
+  commonStore.setRecordingRawContent([]);
+}
+
+export async function getSoundFont() {
+  let soundUrl: string;
+  if (commonStore.compositionParams.useLocalSoundFont)
+    soundUrl = 'assets/sound-font';
+  else
+    soundUrl = !commonStore.settings.giteeUpdatesSource ?
+      `https://raw.githubusercontent.com/josStorer/sgm_plus/master` :
+      `https://cdn.jsdelivr.net/gh/josstorer/sgm_plus`;
+  const fallbackUrl = 'https://cdn.jsdelivr.net/gh/josstorer/sgm_plus';
+  await fetch(soundUrl + '/soundfont.json').then(r => {
+    if (!r.ok)
+      soundUrl = fallbackUrl;
+  }).catch(() => soundUrl = fallbackUrl);
+  return soundUrl;
+}
+
+export const setActivePreset = (preset: Preset | null) => {
+  commonStore.setActivePreset(preset);
+  //TODO if (preset.displayPresetMessages) {
+  const conversation: Conversation = {};
+  const conversationOrder: string[] = [];
+  if (preset)
+    for (const message of preset.messages) {
+      const newUuid = uuid();
+      conversationOrder.push(newUuid);
+      conversation[newUuid] = {
+        sender: message.role === 'user' ? userName : botName,
+        type: MessageType.Normal,
+        color: message.role === 'user' ? 'brand' : 'colorful',
+        time: new Date().toISOString(),
+        content: message.content,
+        side: message.role === 'user' ? 'right' : 'left',
+        done: true
+      };
+    }
+  commonStore.setConversation(conversation);
+  commonStore.setConversationOrder(conversationOrder);
+  //}
+};
+
 export function getSupportedCustomCudaFile(isBeta: boolean) {
   if ([' 10', ' 16', ' 20', ' 30', 'MX', 'Tesla P', 'Quadro P', 'NVIDIA P', 'TITAN X', 'TITAN RTX', 'RTX A',
     'Quadro RTX 4000', 'Quadro RTX 5000', 'Tesla T4', 'NVIDIA A10', 'NVIDIA A40'].some(v => commonStore.status.device_name.includes(v)))

frontend/src/utils/web-file-operations.ts

@@ -0,0 +1,77 @@
+import { getDocument, GlobalWorkerOptions, PDFDocumentProxy } from 'pdfjs-dist';
+import { TextItem } from 'pdfjs-dist/types/src/display/api';
+
+export function webOpenOpenFileDialog(filterPattern: string, fnStartLoading: Function | undefined): Promise<{
+  blob: Blob,
+  content?: string
+}> {
+  return new Promise((resolve, reject) => {
+    const input = document.createElement('input');
+    input.type = 'file';
+    input.accept = filterPattern
+    .replaceAll('*.txt', 'text/plain')
+    .replace('*.midi', 'audio/midi')
+    .replace('*.mid', 'audio/midi')
+    .replaceAll('*.', 'application/')
+    .replaceAll(';', ',');
+
+    input.onchange = async e => {
+      // @ts-ignore
+      const file: Blob = e.target?.files[0];
+      if (fnStartLoading && typeof fnStartLoading === 'function')
+        fnStartLoading();
+      if (!GlobalWorkerOptions.workerSrc && file.type === 'application/pdf')
+        // @ts-ignore
+        GlobalWorkerOptions.workerSrc = await import('pdfjs-dist/build/pdf.worker.min.mjs');
+      if (file.type === 'text/plain') {
+        const reader = new FileReader();
+        reader.readAsText(file, 'UTF-8');
+
+        reader.onload = event => {
+          const content = event.target?.result as string;
+          resolve({
+            blob: file,
+            content: content
+          });
+        };
+        reader.onerror = reject;
+      } else if (file.type === 'application/pdf') {
+        const readPDFPage = async (doc: PDFDocumentProxy, pageNo: number) => {
+          const page = await doc.getPage(pageNo);
+          const tokenizedText = await page.getTextContent();
+          return tokenizedText.items.map(token => (token as TextItem).str).join('');
+        };
+        let reader = new FileReader();
+        reader.readAsArrayBuffer(file);
+
+        reader.onload = async (event) => {
+          try {
+            const doc = await getDocument(event.target?.result!).promise;
+            const pageTextPromises = [];
+            for (let pageNo = 1; pageNo <= doc.numPages; pageNo++) {
+              pageTextPromises.push(readPDFPage(doc, pageNo));
+            }
+            const pageTexts = await Promise.all(pageTextPromises);
+            let content;
+            if (pageTexts.length === 1)
+              content = pageTexts[0];
+            else
+              content = pageTexts.map((p, i) => `Page ${i + 1}:\n${p}`).join('\n\n');
+            resolve({
+              blob: file,
+              content: content
+            });
+          } catch (err) {
+            reject(err);
+          }
+        };
+        reader.onerror = reject;
+      } else {
+        resolve({
+          blob: file
+        });
+      }
+    };
+    input.click();
+  });
+}

frontend/src/webWails.js

@@ -1,3 +1,5 @@
+import { webOpenOpenFileDialog } from './utils/web-file-operations'
+
 function defineRuntime(name, func) {
   window.runtime[name] = func
 }
@@ -46,6 +48,8 @@ if (!window.go) {
   // not implemented
   defineApp('AddToDownloadList', async () => {
   })
+  defineApp('CloseMidiPort', async () => {
+  })
   defineApp('ContinueDownload', async () => {
   })
   defineApp('ConvertData', async () => {
@@ -72,8 +76,12 @@ if (!window.go) {
   })
   defineApp('OpenFileFolder', async () => {
   })
+  defineApp('OpenMidiPort', async () => {
+  })
   defineApp('PauseDownload', async () => {
   })
+  defineApp('PlayNote', async () => {
+  })
   defineApp('ReadFileInfo', async () => {
   })
   defineApp('RestartApp', async () => {
@@ -107,37 +115,7 @@ if (!window.go) {
   defineApp('ListDirFiles', async () => {
     return []
   })
-  defineApp('OpenOpenFileDialog', async (filterPattern) => {
-    return new Promise((resolve, reject) => {
-      const input = document.createElement('input')
-      input.type = 'file'
-      input.accept = filterPattern
-        .replaceAll('*.txt', 'text/plain')
-        .replaceAll('*.', 'application/')
-        .replaceAll(';', ',')
-
-      input.onchange = e => {
-        const file = e.target?.files[0]
-        if (file.type === 'text/plain') {
-          const reader = new FileReader()
-          reader.readAsText(file, 'UTF-8')
-
-          reader.onload = readerEvent => {
-            const content = readerEvent.target?.result
-            resolve({
-              blob: file,
-              content: content
-            })
-          }
-        } else {
-          resolve({
-            blob: file
-          })
-        }
-      }
-      input.click()
-    })
-  })
+  defineApp('OpenOpenFileDialog', webOpenOpenFileDialog)
   defineApp('OpenSaveFileDialog', async (filterPattern, defaultFileName, savedContent) => {
     const saver = await import('file-saver')
     saver.saveAs(new Blob([savedContent], { type: 'text/plain;charset=utf-8' }), defaultFileName)

frontend/vite.config.ts

@@ -3,6 +3,7 @@ import { dependencies } from './package.json';
 import { defineConfig } from 'vite';
 import react from '@vitejs/plugin-react';
 import { visualizer } from 'rollup-plugin-visualizer';
+import topLevelAwait from 'vite-plugin-top-level-await';
 
 // dependencies that exist anywhere
 const vendor = [
@@ -20,6 +21,7 @@ const embedded = [
 
   // dependencies that exist in single component
   'react-beautiful-dnd',
+  'react-draggable',
   '@magenta/music', 'html-midi-player',
   'react-markdown', 'rehype-highlight', 'rehype-raw', 'remark-breaks', 'remark-gfm'
 ];
@@ -35,12 +37,18 @@ function renderChunks(deps: Record<string, string>) {
 
 // https://vitejs.dev/config/
 export default defineConfig({
-  plugins: [react(),
+  plugins: [
+    react(),
     visualizer({
       template: 'treemap',
       gzipSize: true,
       brotliSize: true
-    })],
+    }),
+    topLevelAwait({
+      promiseExportName: '__tla',
+      promiseImportName: i => `__tla_${i}`
+    })
+  ],
   build: {
     chunkSizeWarningLimit: 3000,
     rollupOptions: {
@@ -48,7 +56,10 @@ export default defineConfig({
         manualChunks: {
           vendor,
           ...renderChunks(dependencies)
-        }
+        },
+        entryFileNames: `assets/[name].js`,
+        chunkFileNames: `assets/[name].js`,
+        assetFileNames: `assets/[name].[ext]`
       }
     }
   }

frontend/wailsjs/go/backend_golang/App.d.ts

@@ -4,13 +4,19 @@ import {backend_golang} from '../models';
 
 export function AddToDownloadList(arg1:string,arg2:string):Promise<void>;
 
+export function CloseMidiPort():Promise<void>;
+
 export function ContinueDownload(arg1:string):Promise<void>;
 
 export function ConvertData(arg1:string,arg2:string,arg3:string,arg4:string):Promise<string>;
 
+export function ConvertGGML(arg1:string,arg2:string,arg3:string,arg4:boolean):Promise<string>;
+
 export function ConvertModel(arg1:string,arg2:string,arg3:string,arg4:string):Promise<string>;
 
-export function ConvertSafetensors(arg1:string,arg2:string,arg3:string):Promise<string>;
+export function ConvertSafetensors(arg1:string,arg2:string):Promise<string>;
+
+export function ConvertSafetensorsWithPython(arg1:string,arg2:string,arg3:string):Promise<string>;
 
 export function CopyFile(arg1:string,arg2:string):Promise<void>;
 
@@ -36,6 +42,8 @@ export function MergeLora(arg1:string,arg2:boolean,arg3:number,arg4:string,arg5:
 
 export function OpenFileFolder(arg1:string,arg2:boolean):Promise<void>;
 
+export function OpenMidiPort(arg1:number):Promise<void>;
+
 export function OpenOpenFileDialog(arg1:string):Promise<string>;
 
 export function OpenSaveFileDialog(arg1:string,arg2:string,arg3:string):Promise<string>;
@@ -44,15 +52,21 @@ export function OpenSaveFileDialogBytes(arg1:string,arg2:string,arg3:Array<numbe
 
 export function PauseDownload(arg1:string):Promise<void>;
 
+export function PlayNote(arg1:backend_golang.MIDIMessage):Promise<void>;
+
 export function ReadFileInfo(arg1:string):Promise<backend_golang.FileInfo>;
 
 export function ReadJson(arg1:string):Promise<any>;
 
 export function RestartApp():Promise<void>;
 
+export function SaveFile(arg1:string,arg2:Array<number>):Promise<void>;
+
 export function SaveJson(arg1:string,arg2:any):Promise<void>;
 
-export function StartServer(arg1:string,arg2:number,arg3:string,arg4:boolean,arg5:boolean):Promise<string>;
+export function StartFile(arg1:string):Promise<void>;
+
+export function StartServer(arg1:string,arg2:number,arg3:string,arg4:boolean,arg5:boolean,arg6:boolean,arg7:boolean):Promise<string>;
 
 export function StartWebGPUServer(arg1:number,arg2:string):Promise<string>;
 

frontend/wailsjs/go/backend_golang/App.js

@@ -6,6 +6,10 @@ export function AddToDownloadList(arg1, arg2) {
   return window['go']['backend_golang']['App']['AddToDownloadList'](arg1, arg2);
 }
 
+export function CloseMidiPort() {
+  return window['go']['backend_golang']['App']['CloseMidiPort']();
+}
+
 export function ContinueDownload(arg1) {
   return window['go']['backend_golang']['App']['ContinueDownload'](arg1);
 }
@@ -14,12 +18,20 @@ export function ConvertData(arg1, arg2, arg3, arg4) {
   return window['go']['backend_golang']['App']['ConvertData'](arg1, arg2, arg3, arg4);
 }
 
+export function ConvertGGML(arg1, arg2, arg3, arg4) {
+  return window['go']['backend_golang']['App']['ConvertGGML'](arg1, arg2, arg3, arg4);
+}
+
 export function ConvertModel(arg1, arg2, arg3, arg4) {
   return window['go']['backend_golang']['App']['ConvertModel'](arg1, arg2, arg3, arg4);
 }
 
-export function ConvertSafetensors(arg1, arg2, arg3) {
-  return window['go']['backend_golang']['App']['ConvertSafetensors'](arg1, arg2, arg3);
+export function ConvertSafetensors(arg1, arg2) {
+  return window['go']['backend_golang']['App']['ConvertSafetensors'](arg1, arg2);
+}
+
+export function ConvertSafetensorsWithPython(arg1, arg2, arg3) {
+  return window['go']['backend_golang']['App']['ConvertSafetensorsWithPython'](arg1, arg2, arg3);
 }
 
 export function CopyFile(arg1, arg2) {
@@ -70,6 +82,10 @@ export function OpenFileFolder(arg1, arg2) {
   return window['go']['backend_golang']['App']['OpenFileFolder'](arg1, arg2);
 }
 
+export function OpenMidiPort(arg1) {
+  return window['go']['backend_golang']['App']['OpenMidiPort'](arg1);
+}
+
 export function OpenOpenFileDialog(arg1) {
   return window['go']['backend_golang']['App']['OpenOpenFileDialog'](arg1);
 }
@@ -86,6 +102,10 @@ export function PauseDownload(arg1) {
   return window['go']['backend_golang']['App']['PauseDownload'](arg1);
 }
 
+export function PlayNote(arg1) {
+  return window['go']['backend_golang']['App']['PlayNote'](arg1);
+}
+
 export function ReadFileInfo(arg1) {
   return window['go']['backend_golang']['App']['ReadFileInfo'](arg1);
 }
@@ -98,12 +118,20 @@ export function RestartApp() {
   return window['go']['backend_golang']['App']['RestartApp']();
 }
 
+export function SaveFile(arg1, arg2) {
+  return window['go']['backend_golang']['App']['SaveFile'](arg1, arg2);
+}
+
 export function SaveJson(arg1, arg2) {
   return window['go']['backend_golang']['App']['SaveJson'](arg1, arg2);
 }
 
-export function StartServer(arg1, arg2, arg3, arg4, arg5) {
-  return window['go']['backend_golang']['App']['StartServer'](arg1, arg2, arg3, arg4, arg5);
+export function StartFile(arg1) {
+  return window['go']['backend_golang']['App']['StartFile'](arg1);
+}
+
+export function StartServer(arg1, arg2, arg3, arg4, arg5, arg6, arg7) {
+  return window['go']['backend_golang']['App']['StartServer'](arg1, arg2, arg3, arg4, arg5, arg6, arg7);
 }
 
 export function StartWebGPUServer(arg1, arg2) {

frontend/wailsjs/go/models.ts

@@ -18,6 +18,28 @@ export namespace backend_golang {
 	        this.modTime = source["modTime"];
 	    }
 	}
+	export class MIDIMessage {
+	    messageType: string;
+	    channel: number;
+	    note: number;
+	    velocity: number;
+	    control: number;
+	    value: number;
+	
+	    static createFrom(source: any = {}) {
+	        return new MIDIMessage(source);
+	    }
+	
+	    constructor(source: any = {}) {
+	        if ('string' === typeof source) source = JSON.parse(source);
+	        this.messageType = source["messageType"];
+	        this.channel = source["channel"];
+	        this.note = source["note"];
+	        this.velocity = source["velocity"];
+	        this.control = source["control"];
+	        this.value = source["value"];
+	    }
+	}
 
 }
 

go.mod

@@ -5,16 +5,18 @@ go 1.20
 require (
 	github.com/cavaliergopher/grab/v3 v3.0.1
 	github.com/fsnotify/fsnotify v1.6.0
+	github.com/mattrtaylor/go-rtmidi v0.0.0-20220428034745-af795b1c1a79
 	github.com/minio/selfupdate v0.6.0
 	github.com/nyaosorg/go-windows-su v0.2.1
 	github.com/ubuntu/gowsl v0.0.0-20230615094051-94945650cc1e
-	github.com/wailsapp/wails/v2 v2.6.0
+	github.com/wailsapp/wails/v2 v2.7.1
 )
 
 require (
 	aead.dev/minisign v0.2.0 // indirect
 	github.com/bep/debounce v1.2.1 // indirect
 	github.com/go-ole/go-ole v1.2.6 // indirect
+	github.com/godbus/dbus/v5 v5.1.0 // indirect
 	github.com/google/uuid v1.3.0 // indirect
 	github.com/jchv/go-winloader v0.0.0-20210711035445-715c2860da7e // indirect
 	github.com/labstack/echo/v4 v4.10.2 // indirect
@@ -22,6 +24,7 @@ require (
 	github.com/leaanthony/go-ansi-parser v1.6.0 // indirect
 	github.com/leaanthony/gosod v1.0.3 // indirect
 	github.com/leaanthony/slicer v1.6.0 // indirect
+	github.com/leaanthony/u v1.1.0 // indirect
 	github.com/mattn/go-colorable v0.1.13 // indirect
 	github.com/mattn/go-isatty v0.0.19 // indirect
 	github.com/pkg/browser v0.0.0-20210911075715-681adbf594b8 // indirect
@@ -33,11 +36,11 @@ require (
 	github.com/ubuntu/decorate v0.0.0-20230125165522-2d5b0a9bb117 // indirect
 	github.com/valyala/bytebufferpool v1.0.0 // indirect
 	github.com/valyala/fasttemplate v1.2.2 // indirect
-	github.com/wailsapp/go-webview2 v1.0.1 // indirect
+	github.com/wailsapp/go-webview2 v1.0.10 // indirect
 	github.com/wailsapp/mimetype v1.4.1 // indirect
-	golang.org/x/crypto v0.9.0 // indirect
+	golang.org/x/crypto v0.14.0 // indirect
 	golang.org/x/exp v0.0.0-20230522175609-2e198f4a06a1 // indirect
-	golang.org/x/net v0.10.0 // indirect
-	golang.org/x/sys v0.9.0 // indirect
-	golang.org/x/text v0.9.0 // indirect
+	golang.org/x/net v0.17.0 // indirect
+	golang.org/x/sys v0.13.0 // indirect
+	golang.org/x/text v0.13.0 // indirect
 )

go.sum

@@ -12,6 +12,8 @@ github.com/fsnotify/fsnotify v1.6.0 h1:n+5WquG0fcWoWp6xPWfHdbskMCQaFnG6PfBrh1Ky4
 github.com/fsnotify/fsnotify v1.6.0/go.mod h1:sl3t1tCWJFWoRz9R8WJCbQihKKwmorjAbSClcnxKAGw=
 github.com/go-ole/go-ole v1.2.6 h1:/Fpf6oFPoeFik9ty7siob0G6Ke8QvQEuVcuChpwXzpY=
 github.com/go-ole/go-ole v1.2.6/go.mod h1:pprOEPIfldk/42T2oK7lQ4v4JSDwmV0As9GaiUsvbm0=
+github.com/godbus/dbus/v5 v5.1.0 h1:4KLkAxT3aOY8Li4FRJe/KvhoNFFxo0m6fNuFUO8QJUk=
+github.com/godbus/dbus/v5 v5.1.0/go.mod h1:xhWf0FNVPg57R7Z0UbKHbJfkEywrmjJnf7w5xrFpKfA=
 github.com/google/uuid v1.3.0 h1:t6JiXgmwXMjEs8VusXIJk2BXHsn+wx8BZdTaoZ5fu7I=
 github.com/google/uuid v1.3.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
 github.com/jchv/go-winloader v0.0.0-20210711035445-715c2860da7e h1:Q3+PugElBCf4PFpxhErSzU3/PY5sFL5Z6rfv4AbGAck=
@@ -29,6 +31,8 @@ github.com/leaanthony/gosod v1.0.3/go.mod h1:BJ2J+oHsQIyIQpnLPjnqFGTMnOZXDbvWtRC
 github.com/leaanthony/slicer v1.5.0/go.mod h1:FwrApmf8gOrpzEWM2J/9Lh79tyq8KTX5AzRtwV7m4AY=
 github.com/leaanthony/slicer v1.6.0 h1:1RFP5uiPJvT93TAHi+ipd3NACobkW53yUiBqZheE/Js=
 github.com/leaanthony/slicer v1.6.0/go.mod h1:o/Iz29g7LN0GqH3aMjWAe90381nyZlDNquK+mtH2Fj8=
+github.com/leaanthony/u v1.1.0 h1:2n0d2BwPVXSUq5yhe8lJPHdxevE2qK5G99PMStMZMaI=
+github.com/leaanthony/u v1.1.0/go.mod h1:9+o6hejoRljvZ3BzdYlVL0JYCwtnAsVuN9pVTQcaRfI=
 github.com/matryer/is v1.4.0 h1:sosSmIWwkYITGrxZ25ULNDeKiMNzFSr4V/eqBQP0PeE=
 github.com/matryer/is v1.4.0/go.mod h1:8I/i5uYgLzgsgEloJE1U6xx5HkBQpAZvepWuujKwMRU=
 github.com/mattn/go-colorable v0.1.11/go.mod h1:u5H1YNBxpqRaxsYJYSkiCWKzEfiAb1Gb520KVy5xxl4=
@@ -38,6 +42,8 @@ github.com/mattn/go-isatty v0.0.14/go.mod h1:7GGIvUiUoEMVVmxf/4nioHXj79iQHKdU27k
 github.com/mattn/go-isatty v0.0.16/go.mod h1:kYGgaQfpe5nmfYZH+SKPsOc2e4SrIfOl2e/yFXSvRLM=
 github.com/mattn/go-isatty v0.0.19 h1:JITubQf0MOLdlGRuRq+jtsDlekdYPia9ZFsB8h/APPA=
 github.com/mattn/go-isatty v0.0.19/go.mod h1:W+V8PltTTMOvKvAeJH7IuucS94S2C6jfK/D7dTCTo3Y=
+github.com/mattrtaylor/go-rtmidi v0.0.0-20220428034745-af795b1c1a79 h1:CA1UHN3RuY70DlC0RlvgtB1e8h3kYzmvK7s8CFe+Ohw=
+github.com/mattrtaylor/go-rtmidi v0.0.0-20220428034745-af795b1c1a79/go.mod h1:oBuZjmjlKSj9CZKrNhcx/adNhHiiE0hZknECjIP8Z0Q=
 github.com/minio/selfupdate v0.6.0 h1:i76PgT0K5xO9+hjzKcacQtO7+MjJ4JKA8Ak8XQ9DDwU=
 github.com/minio/selfupdate v0.6.0/go.mod h1:bO02GTIPCMQFTEvE5h4DjYB58bCoZ35XLeBf0buTDdM=
 github.com/nyaosorg/go-windows-su v0.2.1 h1:5V0XavLyjOqPUp7psxxCvBISaneU4XmFPSMlejSl5sc=
@@ -69,24 +75,24 @@ github.com/valyala/bytebufferpool v1.0.0/go.mod h1:6bBcMArwyJ5K/AmCkWv1jt77kVWyC
 github.com/valyala/fasttemplate v1.2.1/go.mod h1:KHLXt3tVN2HBp8eijSv/kGJopbvo7S+qRAEEKiv+SiQ=
 github.com/valyala/fasttemplate v1.2.2 h1:lxLXG0uE3Qnshl9QyaK6XJxMXlQZELvChBOCmQD0Loo=
 github.com/valyala/fasttemplate v1.2.2/go.mod h1:KHLXt3tVN2HBp8eijSv/kGJopbvo7S+qRAEEKiv+SiQ=
-github.com/wailsapp/go-webview2 v1.0.1 h1:dEJIeEApW/MhO2tTMISZBFZPuW7kwrFA1NtgFB1z1II=
-github.com/wailsapp/go-webview2 v1.0.1/go.mod h1:Uk2BePfCRzttBBjFrBmqKGJd41P6QIHeV9kTgIeOZNo=
+github.com/wailsapp/go-webview2 v1.0.10 h1:PP5Hug6pnQEAhfRzLCoOh2jJaPdrqeRgJKZhyYyDV/w=
+github.com/wailsapp/go-webview2 v1.0.10/go.mod h1:Uk2BePfCRzttBBjFrBmqKGJd41P6QIHeV9kTgIeOZNo=
 github.com/wailsapp/mimetype v1.4.1 h1:pQN9ycO7uo4vsUUuPeHEYoUkLVkaRntMnHJxVwYhwHs=
 github.com/wailsapp/mimetype v1.4.1/go.mod h1:9aV5k31bBOv5z6u+QP8TltzvNGJPmNJD4XlAL3U+j3o=
-github.com/wailsapp/wails/v2 v2.6.0 h1:EyH0zR/EO6dDiqNy8qU5spaXDfkluiq77xrkabPYD4c=
-github.com/wailsapp/wails/v2 v2.6.0/go.mod h1:WBG9KKWuw0FKfoepBrr/vRlyTmHaMibWesK3yz6nNiM=
+github.com/wailsapp/wails/v2 v2.7.1 h1:HAzp2c5ODOzsLC6ZMDVtNOB72ozM7/SJecJPB2Ur+UU=
+github.com/wailsapp/wails/v2 v2.7.1/go.mod h1:oIJVwwso5fdOgprBYWXBBqtx6PaSvxg8/KTQHNGkadc=
 golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w=
 golang.org/x/crypto v0.0.0-20210220033148-5ea612d1eb83/go.mod h1:jdWPYTVW3xRLrWPugEBEK3UY2ZEsg3UU495nc5E+M+I=
 golang.org/x/crypto v0.0.0-20211209193657-4570a0811e8b/go.mod h1:IxCIyHEi3zRg3s0A5j5BB6A9Jmi73HwBIUl50j+osU4=
-golang.org/x/crypto v0.9.0 h1:LF6fAI+IutBocDJ2OT0Q1g8plpYljMZ4+lty+dsqw3g=
-golang.org/x/crypto v0.9.0/go.mod h1:yrmDGqONDYtNj3tH8X9dzUun2m2lzPa9ngI6/RUPGR0=
+golang.org/x/crypto v0.14.0 h1:wBqGXzWJW6m1XrIKlAH0Hs1JJ7+9KBwnIO8v66Q9cHc=
+golang.org/x/crypto v0.14.0/go.mod h1:MVFd36DqK4CsrnJYDkBA3VC4m2GkXAM0PvzMCn4JQf4=
 golang.org/x/exp v0.0.0-20230522175609-2e198f4a06a1 h1:k/i9J1pBpvlfR+9QsetwPyERsqu1GIbi967PQMq3Ivc=
 golang.org/x/exp v0.0.0-20230522175609-2e198f4a06a1/go.mod h1:V1LtkGg67GoY2N1AnLN78QLrzxkLyJw7RJb1gzOOz9w=
 golang.org/x/net v0.0.0-20190404232315-eb5bcb51f2a3/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
 golang.org/x/net v0.0.0-20210505024714-0287a6fb4125/go.mod h1:9nx3DQGgdP8bBQD5qxJ1jj9UTztislL4KSBs9R2vV5Y=
 golang.org/x/net v0.0.0-20211112202133-69e39bad7dc2/go.mod h1:9nx3DQGgdP8bBQD5qxJ1jj9UTztislL4KSBs9R2vV5Y=
-golang.org/x/net v0.10.0 h1:X2//UzNDwYmtCLn7To6G58Wr6f5ahEAQgKNzv9Y951M=
-golang.org/x/net v0.10.0/go.mod h1:0qNGK6F8kojg2nk9dLZ2mShWaEBan6FAoqfSigmmuDg=
+golang.org/x/net v0.17.0 h1:pVaXccu2ozPjCXewfr1S7xza/zcXTity9cCdXQYSjIM=
+golang.org/x/net v0.17.0/go.mod h1:NxSsAGuq816PNPmqtQdLE42eU2Fs7NoRIZrHJAlaCOE=
 golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20190916202348-b4ddaad3f8a3/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20191026070338-33540a1f6037/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
@@ -103,14 +109,14 @@ golang.org/x/sys v0.0.0-20220715151400-c0bba94af5f8/go.mod h1:oPkhp1MJrh7nUepCBc
 golang.org/x/sys v0.0.0-20220811171246-fbc7d0a398ab/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.0.0-20220908164124-27713097b956/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.9.0 h1:KS/R3tvhPqvJvwcKfnBHJwwthS11LRhmM5D59eEXa0s=
-golang.org/x/sys v0.9.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.13.0 h1:Af8nKPmuFypiUBjVoU9V20FiaFXOcuZI21p0ycVYYGE=
+golang.org/x/sys v0.13.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/term v0.0.0-20201117132131-f5c789dd3221/go.mod h1:Nr5EML6q2oocZ2LXRh80K7BxOlk5/8JxuGnuhpl+muw=
 golang.org/x/term v0.0.0-20201126162022-7de9c90e9dd1/go.mod h1:bj7SfCRtBDWHUb9snDiAeCFNEtKQo2Wmx5Cou7ajbmo=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
 golang.org/x/text v0.3.6/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
-golang.org/x/text v0.9.0 h1:2sjJmO8cDvYveuX97RDLsxlyUxLl+GHoLxBiRdHllBE=
-golang.org/x/text v0.9.0/go.mod h1:e1OnstbJyHTd6l/uOt8jFFHp6TRDWZR/bV3emEE/zU8=
+golang.org/x/text v0.13.0 h1:ablQoSUd0tRdKxZewP80B+BaqeKJuVhuRxj/dkrun3k=
+golang.org/x/text v0.13.0/go.mod h1:TvPlkZtksWOMsz7fbANvkp4WM8x/WCo/om8BMLbz+aE=
 golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=

main.go

@@ -11,6 +11,7 @@ import (
 	backend "rwkv-runner/backend-golang"
 
 	"github.com/wailsapp/wails/v2"
+	wailsLogger "github.com/wailsapp/wails/v2/pkg/logger"
 	"github.com/wailsapp/wails/v2/pkg/options"
 	"github.com/wailsapp/wails/v2/pkg/options/assetserver"
 	"github.com/wailsapp/wails/v2/pkg/options/windows"
@@ -66,7 +67,10 @@ var midiAssets embed.FS
 var components embed.FS
 
 func main() {
+	dev := true
 	if buildInfo, ok := debug.ReadBuildInfo(); !ok || strings.Contains(buildInfo.String(), "-ldflags") {
+		dev = false
+
 		backend.CopyEmbed(assets)
 		os.RemoveAll("./py310/Lib/site-packages/cyac-1.7.dist-info")
 		backend.CopyEmbed(cyac)
@@ -94,11 +98,18 @@ func main() {
 		app.HasConfigData = false
 	}
 
+	var logger wailsLogger.Logger
+	if dev {
+		logger = wailsLogger.NewDefaultLogger()
+	} else {
+		logger = wailsLogger.NewFileLogger("crash.log")
+	}
+
 	// Create application with options
 	err = wails.Run(&options.App{
 		Title:                    "RWKV-Runner",
 		Width:                    1024,
-		Height:                   680,
+		Height:                   700,
 		MinWidth:                 375,
 		MinHeight:                640,
 		EnableDefaultContextMenu: true,
@@ -115,6 +126,7 @@ func main() {
 		Bind: []any{
 			app,
 		},
+		Logger: logger,
 	})
 
 	if err != nil {

manifest.json

@@ -1,5 +1,5 @@
 {
-  "version": "1.5.1",
+  "version": "1.6.3",
   "introduction": {
     "en": "RWKV is an open-source, commercially usable large language model with high flexibility and great potential for development.\n### About This Tool\nThis tool aims to lower the barrier of entry for using large language models, making it accessible to everyone. It provides fully automated dependency and model management. You simply need to click and run, following the instructions, to deploy a local large language model. The tool itself is very compact and only requires a single executable file for one-click deployment.\nAdditionally, this tool offers an interface that is fully compatible with the OpenAI API. This means you can use any ChatGPT client as a client for RWKV, enabling capability expansion beyond just chat functionality.\n### Preset Configuration Rules at the Bottom\nThis tool comes with a series of preset configurations to reduce complexity. The naming rules for each configuration represent the following in order: device - required VRAM/memory - model size - model language.\nFor example, \"GPU-8G-3B-EN\" indicates that this configuration is for a graphics card with 8GB of VRAM, a model size of 3 billion parameters, and it uses an English language model.\nLarger model sizes have higher performance and VRAM requirements. Among configurations with the same model size, those with higher VRAM usage will have faster runtime.\nFor example, if you have 12GB of VRAM but running the \"GPU-12G-7B-EN\" configuration is slow, you can downgrade to \"GPU-8G-3B-EN\" for a significant speed improvement.\n### About RWKV\nRWKV is an RNN with Transformer-level LLM performance, which can also be directly trained like a GPT transformer (parallelizable). And it's 100% attention-free. You only need the hidden state at position t to compute the state at position t+1. You can use the \"GPT\" mode to quickly compute the hidden state for the \"RNN\" mode.<br/>So it's combining the best of RNN and transformer - great performance, fast inference, saves VRAM, fast training, \"infinite\" ctx_len, and free sentence embedding (using the final hidden state).",
     "zh": "RWKV是一个开源且允许商用的大语言模型，灵活性很高且极具发展潜力。\n### 关于本工具\n本工具旨在降低大语言模型的使用门槛，做到人人可用，本工具提供了全自动化的依赖和模型管理，你只需要直接点击运行，跟随引导，即可完成本地大语言模型的部署，工具本身体积极小，只需要一个exe即可完成一键部署。\n此外，本工具提供了与OpenAI API完全兼容的接口，这意味着你可以把任意ChatGPT客户端用作RWKV的客户端，实现能力拓展，而不局限于聊天。\n### 底部的预设配置规则\n本工具内置了一系列预设配置，以降低使用难度，每个配置名的规则，依次代表着：设备-所需显存/内存-模型规模-模型语言。\n例如，GPU-8G-3B-CN，表示该配置用于显卡，需要8G显存，模型规模为30亿参数，使用的是中文模型。\n模型规模越大，性能要求越高，显存要求也越高，而同样模型规模的配置中，显存占用越高的，运行速度越快。\n例如当你有12G显存，但运行GPU-12G-7B-CN配置速度比较慢，可降级成GPU-8G-3B-CN，将会大幅提速。\n### 关于RWKV\nRWKV是具有Transformer级别LLM性能的RNN，也可以像GPT Transformer一样直接进行训练（可并行化）。而且它是100% attention-free的。你只需在位置t处获得隐藏状态即可计算位置t + 1处的状态。你可以使用“GPT”模式快速计算用于“RNN”模式的隐藏状态。\n因此，它将RNN和Transformer的优点结合起来 - 高性能、快速推理、节省显存、快速训练、“无限”上下文长度以及免费的语句嵌入（使用最终隐藏状态）。"
@@ -15,6 +15,42 @@
     }
   ],
   "models": [
+    {
+      "name": "RWKV-5-World-0.1B-v1-20230803-ctx4096.pth",
+      "desc": {
+        "en": "RWKV-5 Global Languages 0.1B v1",
+        "zh": "RWKV-5 全球语言 0.1B v1",
+        "ja": "RWKV-5 グローバル言語 0.1B v1"
+      },
+      "size": 385598386,
+      "SHA256": "c844a3ee05bcb9065848cb05b10c48a3f381f5ac1953aad89e156ecdf31d7703",
+      "lastUpdated": "2023-08-03T15:18:46",
+      "url": "https://huggingface.co/BlinkDL/rwkv-5-world/blob/main/RWKV-5-World-0.1B-v1-20230803-ctx4096.pth?download=true",
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-5-world/resolve/main/RWKV-5-World-0.1B-v1-20230803-ctx4096.pth?download=true",
+      "tags": [
+        "Main",
+        "RWKV-5",
+        "Global"
+      ]
+    },
+    {
+      "name": "RWKV-5-World-0.4B-v2-20231113-ctx4096.pth",
+      "desc": {
+        "en": "RWKV-5 Global Languages 0.4B v2",
+        "zh": "RWKV-5 全球语言 0.4B v2",
+        "ja": "RWKV-5 グローバル言語 0.4B v2"
+      },
+      "size": 923523954,
+      "SHA256": "5a288c54c7f30b0e2d4af23991133fad2af2d5e59ec7ad850ffe78054a5e4f92",
+      "lastUpdated": "2023-11-14T01:23:49",
+      "url": "https://huggingface.co/BlinkDL/rwkv-5-world/blob/main/RWKV-5-World-0.4B-v2-20231113-ctx4096.pth?download=true",
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-5-world/resolve/main/RWKV-5-World-0.4B-v2-20231113-ctx4096.pth?download=true",
+      "tags": [
+        "Main",
+        "RWKV-5",
+        "Global"
+      ]
+    },
     {
       "name": "RWKV-5-World-1B5-v2-20231025-ctx4096.pth",
       "desc": {
@@ -26,7 +62,48 @@
       "SHA256": "5a89f56be7f82ab9dd0835af9a6838f788477471616c02f7b041e3aea0c57435",
       "lastUpdated": "2023-10-26T05:49:30",
       "url": "https://huggingface.co/BlinkDL/rwkv-5-world/blob/main/RWKV-5-World-1B5-v2-20231025-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-5-world/resolve/main/RWKV-5-World-1B5-v2-20231025-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-5-world/resolve/main/RWKV-5-World-1B5-v2-20231025-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-5",
+        "Global"
+      ]
+    },
+    {
+      "name": "RWKV-5-World-3B-v2-20231113-ctx4096.pth",
+      "desc": {
+        "en": "RWKV-5 Global Languages 3B v2",
+        "zh": "RWKV-5 全球语言 3B v2",
+        "ja": "RWKV-5 グローバル言語 3B v2"
+      },
+      "size": 6126106674,
+      "SHA256": "a4bd430343c6fd138b85bbc68bb20262d3a2f053ea57dc4b41078269af68ff9c",
+      "lastUpdated": "2023-11-14T01:23:49",
+      "url": "https://huggingface.co/BlinkDL/rwkv-5-world/blob/main/RWKV-5-World-3B-v2-20231113-ctx4096.pth?download=true",
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-5-world/resolve/main/RWKV-5-World-3B-v2-20231113-ctx4096.pth?download=true",
+      "tags": [
+        "Main",
+        "RWKV-5",
+        "Global"
+      ]
+    },
+    {
+      "name": "RWKV-5-World-3B-v2-20231118-ctx16k.pth",
+      "desc": {
+        "en": "RWKV-5 Global Languages 3B v2 Ctx16k",
+        "zh": "RWKV-5 全球语言 3B v2 16k上下文",
+        "ja": "RWKV-5 グローバル言語 3B v2 16kコンテキスト"
+      },
+      "size": 6126106467,
+      "SHA256": "efa5178d1c824b94ef17c6c9a456674e5581a8be832becbda9aba4dc533f88c2",
+      "lastUpdated": "2023-11-19T04:21:04",
+      "url": "https://huggingface.co/BlinkDL/rwkv-5-world/blob/main/RWKV-5-World-3B-v2-20231118-ctx16k.pth?download=true",
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-5-world/resolve/main/RWKV-5-World-3B-v2-20231118-ctx16k.pth?download=true",
+      "tags": [
+        "Main",
+        "RWKV-5",
+        "Global"
+      ]
     },
     {
       "name": "RWKV-4-World-CHNtuned-0.1B-v1-20230617-ctx4096.pth",
@@ -39,7 +116,12 @@
       "SHA256": "a3888f9958d378ee6d4976ae1c02edb698f4382e426086febafb4a69417b9080",
       "lastUpdated": "2023-06-17T18:35:26",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-world/blob/main/RWKV-4-World-CHNtuned-0.1B-v1-20230617-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-CHNtuned-0.1B-v1-20230617-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-CHNtuned-0.1B-v1-20230617-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "CN"
+      ]
     },
     {
       "name": "RWKV-4-World-0.1B-v1-20230520-ctx4096.pth",
@@ -52,7 +134,12 @@
       "SHA256": "a10ef99df2a8f8a6801edf4fc92a9c49bedd63dcb900d3e5667a2136b3d671e7",
       "lastUpdated": "2023-05-25T09:21:27",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-world/blob/main/RWKV-4-World-0.1B-v1-20230520-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-0.1B-v1-20230520-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-0.1B-v1-20230520-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Global"
+      ]
     },
     {
       "name": "RWKV-4-World-CHNtuned-0.4B-v1-20230618-ctx4096.pth",
@@ -65,7 +152,12 @@
       "SHA256": "dbd5302cbee596bbc900f97eb10b2af3001a7f2c7e4d8643bf8683b2cdbdd324",
       "lastUpdated": "2023-06-18T10:46:50",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-world/blob/main/RWKV-4-World-CHNtuned-0.4B-v1-20230618-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-CHNtuned-0.4B-v1-20230618-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-CHNtuned-0.4B-v1-20230618-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "CN"
+      ]
     },
     {
       "name": "RWKV-4-World-0.4B-v1-20230529-ctx4096.pth",
@@ -78,7 +170,12 @@
       "SHA256": "4b4a2733cf5e5dc97dd62106f391d99895d16b11c5ccd10c89f28c52067a4919",
       "lastUpdated": "2023-05-29T13:25:53",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-world/blob/main/RWKV-4-World-0.4B-v1-20230529-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-0.4B-v1-20230529-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-0.4B-v1-20230529-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Global"
+      ]
     },
     {
       "name": "RWKV-4-World-CHNtuned-1.5B-v1-20230620-ctx4096.pth",
@@ -91,7 +188,12 @@
       "SHA256": "9f31f2ed5fe52dcf2d50208eb2efd764b9674dba2adb1baeff61997b4390a26b",
       "lastUpdated": "2023-06-20T06:35:37",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-world/blob/main/RWKV-4-World-CHNtuned-1.5B-v1-20230620-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-CHNtuned-1.5B-v1-20230620-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-CHNtuned-1.5B-v1-20230620-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "CN"
+      ]
     },
     {
       "name": "RWKV-4-World-1.5B-v1-OnlyForTest_57%_trained-20230529-ctx4096.pth",
@@ -130,6 +232,11 @@
       "lastUpdated": "2023-06-07T09:33:32",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-world/blob/main/RWKV-4-World-1.5B-v1-20230607-ctx4096.pth",
       "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-1.5B-v1-20230607-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Global"
+      ],
       "hide": true
     },
     {
@@ -143,7 +250,12 @@
       "SHA256": "71f0c3229f9227cbcb8ae5fee6461197129a57e26366c4d23a49058417b046c9",
       "lastUpdated": "2023-06-12T06:31:32",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-world/blob/main/RWKV-4-World-1.5B-v1-fixed-20230612-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-1.5B-v1-fixed-20230612-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-1.5B-v1-fixed-20230612-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Global"
+      ]
     },
     {
       "name": "RWKV-4-World-3B-v1-OnlyForTest_35%_trained-20230529-ctx4096.pth",
@@ -208,7 +320,12 @@
       "SHA256": "1b227af317fa25b6939ab3c7cd321226ca48b8fe4bbbd2df3db669f1482c54ba",
       "lastUpdated": "2023-06-20T03:00:51",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-world/blob/main/RWKV-4-World-3B-v1-20230619-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-3B-v1-20230619-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-3B-v1-20230619-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Global"
+      ]
     },
     {
       "name": "RWKV-4-World-CHNtuned-3B-v1-20230625-ctx4096.pth",
@@ -221,7 +338,12 @@
       "SHA256": "7d3b5a4d0e9780a3e3d9ae7c2defbe8564d240bc9a238db4ba70cfb66dc33888",
       "lastUpdated": "2023-06-25T14:53:27",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-world/blob/main/RWKV-4-World-CHNtuned-3B-v1-20230625-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-CHNtuned-3B-v1-20230625-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-CHNtuned-3B-v1-20230625-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "CN"
+      ]
     },
     {
       "name": "RWKV-4-World-7B-v1-OnlyForTest_30%_trained-20230529-ctx4096.pth",
@@ -312,7 +434,12 @@
       "SHA256": "db7b011247a0fe4389e1d76e3d6a904185f85d509c8a44ad18bf401094efc293",
       "lastUpdated": "2023-06-26T16:40:04",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-world/blob/main/RWKV-4-World-7B-v1-20230626-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-7B-v1-20230626-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-7B-v1-20230626-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Global"
+      ]
     },
     {
       "name": "RWKV-claude-4-World-7B-20230805-ctx65k.pth",
@@ -325,7 +452,12 @@
       "SHA256": "8cd25f8a1ab58965993cc47b3b2f99585836eed008a2e44526c258189ea751a6",
       "lastUpdated": "2023-08-05T08:52:20",
       "url": "https://huggingface.co/xiaol/RWKV-claude-4-World-7B-65k/blob/main/RWKV-claude-4-World-7B-20230805-ctx65k.pth",
-      "downloadUrl": "https://huggingface.co/xiaol/RWKV-claude-4-World-7B-65k/resolve/main/RWKV-claude-4-World-7B-20230805-ctx65k.pth"
+      "downloadUrl": "https://huggingface.co/xiaol/RWKV-claude-4-World-7B-65k/resolve/main/RWKV-claude-4-World-7B-20230805-ctx65k.pth",
+      "tags": [
+        "Finetuned",
+        "RWKV-4",
+        "Global"
+      ]
     },
     {
       "name": "RWKV-toolformer-translation-japanese-chinese-english-7B-World-20230815-ctx128k.pth",
@@ -338,7 +470,12 @@
       "SHA256": "648a3b21055bdab77021ce278da80fbada8dcaae0b3d41d1eca9aa194c1fd25f",
       "lastUpdated": "2023-08-15T07:18:23",
       "url": "https://huggingface.co/xiaol/RWKV-toolformer-translation-japanese-chinese-english-7B-World-128k/blob/main/RWKV-toolformer-translation-japanese-chinese-english-7B-World-20230815-ctx128k.pth",
-      "downloadUrl": "https://huggingface.co/xiaol/RWKV-toolformer-translation-japanese-chinese-english-7B-World-128k/resolve/main/RWKV-toolformer-translation-japanese-chinese-english-7B-World-20230815-ctx128k.pth"
+      "downloadUrl": "https://huggingface.co/xiaol/RWKV-toolformer-translation-japanese-chinese-english-7B-World-128k/resolve/main/RWKV-toolformer-translation-japanese-chinese-english-7B-World-20230815-ctx128k.pth",
+      "tags": [
+        "Finetuned",
+        "RWKV-4",
+        "Global"
+      ]
     },
     {
       "name": "RWKV-code-4-World-7B-20230820-ctx32k.pth",
@@ -351,7 +488,12 @@
       "SHA256": "19666620437ae3a5fb06e16a52729d67e449fca155fab3d5861ffe9ecf247404",
       "lastUpdated": "2023-08-20T05:00:17",
       "url": "https://huggingface.co/xiaol/RWKV-Code-7B-world-32k/blob/main/RWKV-code-4-World-7B-20230820-ctx32k.pth",
-      "downloadUrl": "https://huggingface.co/xiaol/RWKV-Code-7B-world-32k/resolve/main/RWKV-code-4-World-7B-20230820-ctx32k.pth"
+      "downloadUrl": "https://huggingface.co/xiaol/RWKV-Code-7B-world-32k/resolve/main/RWKV-code-4-World-7B-20230820-ctx32k.pth",
+      "tags": [
+        "Finetuned",
+        "RWKV-4",
+        "Global"
+      ]
     },
     {
       "name": "wizard-rwkv-4-world-ctx32k.pth",
@@ -364,7 +506,12 @@
       "SHA256": "c5d991f315a1676d4bed93dd91f803b1376096e7a4af5bf72b339d055f53bac7",
       "lastUpdated": "2023-07-29T03:21:47",
       "url": "https://huggingface.co/xiaol/wizard-rwkv-world-7B-ctx32k/blob/main/wizard-rwkv-4-world-ctx32k.pth",
-      "downloadUrl": "https://huggingface.co/xiaol/wizard-rwkv-world-7B-ctx32k/resolve/main/wizard-rwkv-4-world-ctx32k.pth"
+      "downloadUrl": "https://huggingface.co/xiaol/wizard-rwkv-world-7B-ctx32k/resolve/main/wizard-rwkv-4-world-ctx32k.pth",
+      "tags": [
+        "Finetuned",
+        "RWKV-4",
+        "Global"
+      ]
     },
     {
       "name": "RWKV-4-World-CHNtuned-7B-v1-20230709-ctx4096.pth",
@@ -377,7 +524,12 @@
       "SHA256": "52d33e8352a40158d21425fee4f68df1515d6324056f788d2c78a366ef578ffa",
       "lastUpdated": "2023-07-09T18:23:33",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-world/blob/main/RWKV-4-World-CHNtuned-7B-v1-20230709-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-CHNtuned-7B-v1-20230709-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-CHNtuned-7B-v1-20230709-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "CN"
+      ]
     },
     {
       "name": "Readflow-RWKV-4-World-CHNtuned-7B-v1-20230709-ctx32k.pth",
@@ -390,7 +542,12 @@
       "SHA256": "1bd1de8cdbd56b67e1374588fe5d202884049c71278ffcb12f5c4efbdb422ee1",
       "lastUpdated": "2023-07-20T06:11:29",
       "url": "https://huggingface.co/xiaol/readflow-rwkv-4-world-ctx32k/blob/main/Readflow-RWKV-4-World-CHNtuned-7B-v1-20230709-ctx32k.pth",
-      "downloadUrl": "https://huggingface.co/xiaol/readflow-rwkv-4-world-ctx32k/resolve/main/Readflow-RWKV-4-World-CHNtuned-7B-v1-20230709-ctx32k.pth"
+      "downloadUrl": "https://huggingface.co/xiaol/readflow-rwkv-4-world-ctx32k/resolve/main/Readflow-RWKV-4-World-CHNtuned-7B-v1-20230709-ctx32k.pth",
+      "tags": [
+        "Finetuned",
+        "RWKV-4",
+        "CN"
+      ]
     },
     {
       "name": "novel-RWKV-4-World-CHNtuned-7B-v1-20230709-ctx32k.pth",
@@ -403,7 +560,12 @@
       "SHA256": "0fe2415ce61af52a8c38c071b475c01b4c9f8a4f2b4aaed6181f0334f3faf7f4",
       "lastUpdated": "2023-07-28T13:30:59",
       "url": "https://huggingface.co/xiaol/ruotangwx-rwkv-7b-novel-32k/blob/main/novel-RWKV-4-World-CHNtuned-7B-v1-20230709-ctx32k.pth",
-      "downloadUrl": "https://huggingface.co/xiaol/ruotangwx-rwkv-7b-novel-32k/resolve/main/novel-RWKV-4-World-CHNtuned-7B-v1-20230709-ctx32k.pth"
+      "downloadUrl": "https://huggingface.co/xiaol/ruotangwx-rwkv-7b-novel-32k/resolve/main/novel-RWKV-4-World-CHNtuned-7B-v1-20230709-ctx32k.pth",
+      "tags": [
+        "Finetuned",
+        "RWKV-4",
+        "CN"
+      ]
     },
     {
       "name": "chatgal-RWKV-4-World-CHNtuned-7B-v1-20230709-ctx32k-1000.pth",
@@ -416,7 +578,12 @@
       "SHA256": "aaed29cfd1bddee47c48f564aa800eb001f62fd03290d772647d5678e40d66e8",
       "lastUpdated": "2023-07-21T08:59:18",
       "url": "https://huggingface.co/xiaol/chatgal-rwkv-7b-world-32k/blob/main/chatgal-RWKV-4-World-CHNtuned-7B-v1-20230709-ctx32k-1000.pth",
-      "downloadUrl": "https://huggingface.co/xiaol/chatgal-rwkv-7b-world-32k/resolve/main/chatgal-RWKV-4-World-CHNtuned-7B-v1-20230709-ctx32k-1000.pth"
+      "downloadUrl": "https://huggingface.co/xiaol/chatgal-rwkv-7b-world-32k/resolve/main/chatgal-RWKV-4-World-CHNtuned-7B-v1-20230709-ctx32k-1000.pth",
+      "tags": [
+        "Finetuned",
+        "RWKV-4",
+        "CN"
+      ]
     },
     {
       "name": "chatgal-RWKV-4-World-CHNtuned-7B-v1-20230709-ctx32k-500.pth",
@@ -429,7 +596,12 @@
       "SHA256": "b5d347d5dedb4f398ec31489ab87b75b1dee772ae7d0a34c26635cf5d95c8794",
       "lastUpdated": "2023-07-21T07:31:05",
       "url": "https://huggingface.co/xiaol/chatgal-rwkv-7b-world-32k/blob/main/chatgal-RWKV-4-World-CHNtuned-7B-v1-20230709-ctx32k-500.pth",
-      "downloadUrl": "https://huggingface.co/xiaol/chatgal-rwkv-7b-world-32k/resolve/main/chatgal-RWKV-4-World-CHNtuned-7B-v1-20230709-ctx32k-500.pth"
+      "downloadUrl": "https://huggingface.co/xiaol/chatgal-rwkv-7b-world-32k/resolve/main/chatgal-RWKV-4-World-CHNtuned-7B-v1-20230709-ctx32k-500.pth",
+      "tags": [
+        "Finetuned",
+        "RWKV-4",
+        "CN"
+      ]
     },
     {
       "name": "RWKV-4-World-JPNtuned-7B-v1-20230718-ctx4096.pth",
@@ -442,7 +614,12 @@
       "SHA256": "3e4c7664ce893ac1f6bb59cd76664fb5c872cb076bb82dbd534db0555b6e9fa5",
       "lastUpdated": "2023-07-18T20:01:12",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-world/blob/main/RWKV-4-World-JPNtuned-7B-v1-20230718-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-JPNtuned-7B-v1-20230718-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-JPNtuned-7B-v1-20230718-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "JP"
+      ]
     },
     {
       "name": "RWKV-novel-4-World-7B-20230810-ctx128k.pth",
@@ -455,7 +632,12 @@
       "SHA256": "5e429c49e4cab2f29a93f87a80635422c8710d70e5b1d962c078e47d957389c8",
       "lastUpdated": "2023-08-10T06:30:32",
       "url": "https://huggingface.co/xiaol/rwkv-7B-world-novel-128k/blob/main/RWKV-novel-4-World-7B-20230810-ctx128k.pth",
-      "downloadUrl": "https://huggingface.co/xiaol/rwkv-7B-world-novel-128k/resolve/main/RWKV-novel-4-World-7B-20230810-ctx128k.pth"
+      "downloadUrl": "https://huggingface.co/xiaol/rwkv-7B-world-novel-128k/resolve/main/RWKV-novel-4-World-7B-20230810-ctx128k.pth",
+      "tags": [
+        "Finetuned",
+        "RWKV-4",
+        "Global"
+      ]
     },
     {
       "name": "RWKV-4-Novel-7B-v1-ChnEng-ChnPro-20230410-ctx4096.pth",
@@ -467,7 +649,13 @@
       "SHA256": "cd40b661930dea46c0f930c51d99cef6b484fe3d641388981dee5a0c68e2b1c7",
       "lastUpdated": "2023-04-10T13:55:52",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-novel/blob/main/RWKV-4-Novel-7B-v1-ChnEng-ChnPro-20230410-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-novel/resolve/main/RWKV-4-Novel-7B-v1-ChnEng-ChnPro-20230410-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-novel/resolve/main/RWKV-4-Novel-7B-v1-ChnEng-ChnPro-20230410-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Raven",
+        "Global"
+      ]
     },
     {
       "name": "RWKV-4-Novel-3B-v1-ChnEng-20230412-ctx4096.pth",
@@ -479,7 +667,13 @@
       "SHA256": "283c6e6fa10c52a93e9a01d9630f288473267ea152a49c6579b5c0427bdc9c61",
       "lastUpdated": "2023-04-12T13:18:29",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-novel/blob/main/RWKV-4-Novel-3B-v1-ChnEng-20230412-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-novel/resolve/main/RWKV-4-Novel-3B-v1-ChnEng-20230412-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-novel/resolve/main/RWKV-4-Novel-3B-v1-ChnEng-20230412-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Raven",
+        "Global"
+      ]
     },
     {
       "name": "RWKV-4-Novel-7B-v1-ChnEng-20230426-ctx8192.pth",
@@ -491,7 +685,13 @@
       "SHA256": "bd08c75a296bd193dcfadb993fe06d7f9dd91ca3385231f24c592c89d25cd596",
       "lastUpdated": "2023-04-26T18:57:01",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-novel/blob/main/RWKV-4-Novel-7B-v1-ChnEng-20230426-ctx8192.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-novel/resolve/main/RWKV-4-Novel-7B-v1-ChnEng-20230426-ctx8192.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-novel/resolve/main/RWKV-4-Novel-7B-v1-ChnEng-20230426-ctx8192.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Raven",
+        "Global"
+      ]
     },
     {
       "name": "RWKV-4-Novel-3B-v1-Chn-20230412-ctx4096.pth",
@@ -503,7 +703,13 @@
       "SHA256": "c41e0af2cbc66e94121377680e8224a1504fac6c9ea620c395f0a79281db26e7",
       "lastUpdated": "2023-04-12T13:18:29",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-novel/blob/main/RWKV-4-Novel-3B-v1-Chn-20230412-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-novel/resolve/main/RWKV-4-Novel-3B-v1-Chn-20230412-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-novel/resolve/main/RWKV-4-Novel-3B-v1-Chn-20230412-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Raven",
+        "Global"
+      ]
     },
     {
       "name": "RWKV-4-Novel-7B-v1-Chn-20230426-ctx8192.pth",
@@ -515,7 +721,13 @@
       "SHA256": "5fced44febdf80d303250eef9c020f087abded43aaecc8caaea8a9e7f1fb771e",
       "lastUpdated": "2023-04-26T18:57:01",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-novel/blob/main/RWKV-4-Novel-7B-v1-Chn-20230426-ctx8192.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-novel/resolve/main/RWKV-4-Novel-7B-v1-Chn-20230426-ctx8192.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-novel/resolve/main/RWKV-4-Novel-7B-v1-Chn-20230426-ctx8192.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Raven",
+        "Global"
+      ]
     },
     {
       "name": "RWKV-4-Raven-1B5-v11-Eng99%-Other1%-20230425-ctx4096.pth",
@@ -528,6 +740,11 @@
       "lastUpdated": "2023-04-26T14:27:55",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-raven/blob/main/RWKV-4-Raven-1B5-v11-Eng99%25-Other1%25-20230425-ctx4096.pth",
       "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-1B5-v11-Eng99%25-Other1%25-20230425-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Raven"
+      ],
       "hide": true
     },
     {
@@ -540,7 +757,12 @@
       "SHA256": "6bbbffb3ee2372dfa9ef49c599e9a2bc0a01b94b6a264ba9bf5bd524fc38f723",
       "lastUpdated": "2023-05-21T07:08:56",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-raven/blob/main/RWKV-4-Raven-1B5-v12-Eng98%25-Other2%25-20230520-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-1B5-v12-Eng98%25-Other2%25-20230520-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-1B5-v12-Eng98%25-Other2%25-20230520-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Raven"
+      ]
     },
     {
       "name": "RWKV-4-Raven-3B-v11-Eng99%-Other1%-20230425-ctx4096.pth",
@@ -553,6 +775,11 @@
       "lastUpdated": "2023-04-26T14:27:55",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-raven/blob/main/RWKV-4-Raven-3B-v11-Eng99%25-Other1%25-20230425-ctx4096.pth",
       "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-3B-v11-Eng99%25-Other1%25-20230425-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Raven"
+      ],
       "hide": true
     },
     {
@@ -565,7 +792,12 @@
       "SHA256": "1eea1845acfe9729dfdaec66a8d1aeb91a1287d94bebbca5529c13c050540b33",
       "lastUpdated": "2023-05-21T07:13:25",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-raven/blob/main/RWKV-4-Raven-3B-v12-Eng98%25-Other2%25-20230520-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-3B-v12-Eng98%25-Other2%25-20230520-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-3B-v12-Eng98%25-Other2%25-20230520-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Raven"
+      ]
     },
     {
       "name": "RWKV-4-Raven-3B-v11-Eng49%-Chn49%-Jpn1%-Other1%-20230429-ctx4096.pth",
@@ -578,6 +810,12 @@
       "lastUpdated": "2023-04-29T11:51:51",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-raven/blob/main/RWKV-4-Raven-3B-v11-Eng49%25-Chn49%25-Jpn1%25-Other1%25-20230429-ctx4096.pth",
       "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-3B-v11-Eng49%25-Chn49%25-Jpn1%25-Other1%25-20230429-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Raven",
+        "CN"
+      ],
       "hide": true
     },
     {
@@ -590,7 +828,13 @@
       "SHA256": "c0abb4b745ba3523b9d8b3e1293110867ee55b1ef3dc8c122212f78396755721",
       "lastUpdated": "2023-05-28T11:51:12",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-raven/blob/main/RWKV-4-Raven-3B-v12-Eng49%25-Chn49%25-Jpn1%25-Other1%25-20230527-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-3B-v12-Eng49%25-Chn49%25-Jpn1%25-Other1%25-20230527-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-3B-v12-Eng49%25-Chn49%25-Jpn1%25-Other1%25-20230527-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Raven",
+        "CN"
+      ]
     },
     {
       "name": "RWKV-4-Raven-7B-v11x-Eng99%-Other1%-20230429-ctx8192.pth",
@@ -603,6 +847,11 @@
       "lastUpdated": "2023-04-29T11:44:32",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-raven/blob/main/RWKV-4-Raven-7B-v11x-Eng99%25-Other1%25-20230429-ctx8192.pth",
       "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-7B-v11x-Eng99%25-Other1%25-20230429-ctx8192.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Raven"
+      ],
       "hide": true
     },
     {
@@ -615,7 +864,12 @@
       "SHA256": "5a725eaeb9e09b724de6c97e6845dd0283097c7920acd05b46852ab7afa9ec32",
       "lastUpdated": "2023-05-22T10:32:17",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-raven/blob/main/RWKV-4-Raven-7B-v12-Eng98%25-Other2%25-20230521-ctx8192.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-7B-v12-Eng98%25-Other2%25-20230521-ctx8192.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-7B-v12-Eng98%25-Other2%25-20230521-ctx8192.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Raven"
+      ]
     },
     {
       "name": "RWKV-4-Raven-7B-v10x-Eng49%-Chn50%-Other1%-20230423-ctx4096.pth",
@@ -628,6 +882,12 @@
       "lastUpdated": "2023-04-24T07:48:55",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-raven/blob/main/RWKV-4-Raven-7B-v10x-Eng49%25-Chn50%25-Other1%25-20230423-ctx4096.pth",
       "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-7B-v10x-Eng49%25-Chn50%25-Other1%25-20230423-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Raven",
+        "CN"
+      ],
       "hide": true
     },
     {
@@ -641,6 +901,12 @@
       "lastUpdated": "2023-04-30T14:35:59",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-raven/blob/main/RWKV-4-Raven-7B-v11-Eng49%25-Chn49%25-Jpn1%25-Other1%25-20230430-ctx8192.pth",
       "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-7B-v11-Eng49%25-Chn49%25-Jpn1%25-Other1%25-20230430-ctx8192.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Raven",
+        "CN"
+      ],
       "hide": true
     },
     {
@@ -653,7 +919,13 @@
       "SHA256": "6d4a089ff36d5d9d96b669d425fc5e4e3959cab426535b52e2364df08f58b407",
       "lastUpdated": "2023-05-30T23:16:12",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-raven/blob/main/RWKV-4-Raven-7B-v12-Eng49%25-Chn49%25-Jpn1%25-Other1%25-20230530-ctx8192.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-7B-v12-Eng49%25-Chn49%25-Jpn1%25-Other1%25-20230530-ctx8192.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-7B-v12-Eng49%25-Chn49%25-Jpn1%25-Other1%25-20230530-ctx8192.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Raven",
+        "CN"
+      ]
     },
     {
       "name": "RWKV-4-Raven-14B-v11x-Eng99%-Other1%-20230501-ctx8192.pth",
@@ -666,6 +938,11 @@
       "lastUpdated": "2023-05-02T09:43:33",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-raven/blob/main/RWKV-4-Raven-14B-v11x-Eng99%25-Other1%25-20230501-ctx8192.pth",
       "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-14B-v11x-Eng99%25-Other1%25-20230501-ctx8192.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Raven"
+      ],
       "hide": true
     },
     {
@@ -678,7 +955,12 @@
       "SHA256": "1193b5a9ceab572e4dbb9ed1d798eab7bf4793d18904d08bd4bf183579338ae7",
       "lastUpdated": "2023-05-23T11:22:41",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-raven/blob/main/RWKV-4-Raven-14B-v12-Eng98%25-Other2%25-20230523-ctx8192.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-14B-v12-Eng98%25-Other2%25-20230523-ctx8192.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-14B-v12-Eng98%25-Other2%25-20230523-ctx8192.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Raven"
+      ]
     },
     {
       "name": "RWKV-4-MIDI-120M-v1-20230714-ctx4096.pth",
@@ -691,7 +973,12 @@
       "SHA256": "161d27dcf50d0958d230601ba1e0f8e7dd9c236105e92d2b833496412ace430c",
       "lastUpdated": "2023-07-15T08:03:36",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-music/blob/main/RWKV-4-MIDI-120M-v1-20230714-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-music/resolve/main/RWKV-4-MIDI-120M-v1-20230714-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-music/resolve/main/RWKV-4-MIDI-120M-v1-20230714-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Music"
+      ]
     },
     {
       "name": "RWKV-4-MIDI-560M-v1-20230717-ctx4096.pth",
@@ -704,7 +991,12 @@
       "SHA256": "62b21841b24af38ef176e9e9d895d9fff730cea8aa0623f53a1784d74ce828d6",
       "lastUpdated": "2023-07-17T15:02:08",
       "url": "https://huggingface.co/BlinkDL/rwkv-4-music/blob/main/RWKV-4-MIDI-560M-v1-20230717-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-music/resolve/main/RWKV-4-MIDI-560M-v1-20230717-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-4-music/resolve/main/RWKV-4-MIDI-560M-v1-20230717-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-4",
+        "Music"
+      ]
     },
     {
       "name": "RWKV-5-MIDI-120M-v1-20230728-ctx4096.pth",
@@ -717,7 +1009,12 @@
       "SHA256": "c43d4a2ee7a71a331d05d6cd818dd75f7c48c716e4b98c58e4d27231614b0144",
       "lastUpdated": "2023-07-29T02:17:27",
       "url": "https://huggingface.co/BlinkDL/rwkv-5-music/blob/main/RWKV-5-MIDI-120M-v1-20230728-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-5-music/resolve/main/RWKV-5-MIDI-120M-v1-20230728-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-5-music/resolve/main/RWKV-5-MIDI-120M-v1-20230728-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-5",
+        "Music"
+      ]
     },
     {
       "name": "RWKV-5-MIDI-560M-v1-20230902-ctx4096.pth",
@@ -730,7 +1027,12 @@
       "SHA256": "cb4f2fd8956ca8496d6b2e33bff290c2047759b6fe74884903dbf9c73a11cc77",
       "lastUpdated": "2023-09-03T04:48:41",
       "url": "https://huggingface.co/BlinkDL/rwkv-5-music/blob/main/RWKV-5-MIDI-560M-v1-20230902-ctx4096.pth",
-      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-5-music/resolve/main/RWKV-5-MIDI-560M-v1-20230902-ctx4096.pth"
+      "downloadUrl": "https://huggingface.co/BlinkDL/rwkv-5-music/resolve/main/RWKV-5-MIDI-560M-v1-20230902-ctx4096.pth",
+      "tags": [
+        "Main",
+        "RWKV-5",
+        "Music"
+      ]
     }
   ]
 }

vendor.yml

@@ -3,6 +3,7 @@
 - ^backend-python/get-pip\.py
 - ^backend-python/convert_model\.py
 - ^backend-python/convert_safetensors\.py
+- ^backend-python/convert_pytorch_to_ggml\.py linguist-vendored
 - ^backend-python/utils/midi\.py
 - ^build/
 - ^finetune/lora/