release v1.5.3

.github/workflows/release.yml

@@ -134,8 +134,9 @@ jobs:
       - 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
+          rustup target add aarch64-apple-darwin
+          cargo build --release --target aarch64-apple-darwin
+          mv ./target/aarch64-apple-darwin/release/ai00_server ../backend-rust/webgpu_server
           cd ..
           go install github.com/wailsapp/wails/v2/cmd/wails@latest
           rm ./backend-python/rwkv_pip/wkv_cuda.pyd

CURRENT_CHANGE.md

@@ -1,10 +1,19 @@
 ## Changes
 
-### Improvements
+### Changes
 
-- improve mobile webview
-- add client upgrade progress
-- improve user guide
+- add webgpu nf4
+- add sidePanel for Chat page
+- add hf-mirror for cn users
+- use local API when it's working, even if a custom API URL is provided
+- add rwkv API URL Option
+- macos: change default webgpu backend to aarch64-apple-darwin and fix the permission issue
+- strict default_stop
+- fix state_cache of deploy mode
+- fix a finetune bug
+- improve lazy loading ui
+- display lastModelName at the top (WorkHeader)
+- chores
 
 ## Install
 

backend-golang/app.go

@@ -45,7 +45,7 @@ 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.Mkdir(a.exDir+"models", os.ModePerm)
 	os.Mkdir(a.exDir+"lora-models", os.ModePerm)
 	os.Mkdir(a.exDir+"finetune/json2binidx_tool/data", os.ModePerm)

backend-python/routes/completion.py

@@ -35,6 +35,11 @@ default_stop = [
     "\n\nQ",
     "\n\nHuman",
     "\n\nBob",
+    "\n\nAssistant",
+    "\n\nAnswer",
+    "\n\nA",
+    "\n\nBot",
+    "\n\nAlice",
 ]
 
 

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,
             }

backend-python/routes/state_cache.py

@@ -73,12 +73,12 @@ 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")
@@ -157,12 +157,12 @@ 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")
@@ -200,12 +200,12 @@ def longest_prefix_state(body: LongestPrefixStateBody, request: Request):
         }
 
 
-@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")

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

@@ -49,6 +49,10 @@ fi
 echo "loading $loadModel"
 modelInfo=$(python3 ./finetune/get_layer_and_embd.py $loadModel)
 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/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",
@@ -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",
@@ -268,5 +268,8 @@
   "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"
 }

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",
@@ -268,5 +268,8 @@
   "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"
 }

frontend/src/components/ConfigSelector.tsx

@@ -4,7 +4,7 @@ import { Dropdown, Option } 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) => {

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

@@ -11,7 +11,7 @@ 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';
@@ -94,7 +94,7 @@ export const RunButton: FC<{ onClickRun?: MouseEventHandler, iconMode?: boolean
                 navigate({ pathname: '/downloads' });
               },
               { autoClose: 3000 });
-            AddToDownloadList(modelPath, downloadUrl);
+            AddToDownloadList(modelPath, getHfDownloadUrl(downloadUrl));
           } else {
             toast(t('Can not find download url'), { type: 'error' });
           }

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/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';
@@ -26,9 +29,11 @@ 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 { 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';
 
 let chatSseControllers: {
   [id: string]: AbortController
@@ -188,11 +193,125 @@ 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;
@@ -327,8 +446,10 @@ 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 +473,8 @@ 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 +512,164 @@ 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"
+          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;
-                  }
+              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)
+                    if (commonStore.attachmentUploading)
                       return;
 
-                    commonStore.setAttachmentUploading(true);
+                    OpenOpenFileDialog('*.txt;*.pdf').then(async filePath => {
+                      if (!filePath)
+                        return;
 
-                    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'
-                            });
-                          }
-                          commonStore.setAttachmentUploading(false);
-                        }
-                      ).catch(e => {
+                      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,${info.content}`).then(r => r.blob());
+                        blob = await fetch(absPathAsset(filePath)).then(r => r.blob());
+                        attachmentName = filePath.split(/[\\/]/).pop();
+                      }
+                      if (attachmentContent) {
+                        commonStore.setCurrentTempAttachment(
+                          {
+                            name: attachmentName!,
+                            size: blob.size,
+                            content: attachmentContent
+                          });
                         commonStore.setAttachmentUploading(false);
-                        toast(t('Error') + ' - ' + (e.message || e), { type: 'error', autoClose: 2500 });
-                      });
+                      } else {
+                        const urlPath = `/file-to-text?file_name=${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'
+                              });
+                            }
+                            commonStore.setAttachmentUploading(false);
+                          }
+                        ).catch(e => {
+                          commonStore.setAttachmentUploading(false);
+                          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...')
                     }
-                  }).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} (${bytesToReadable(commonStore.currentTempAttachment.size)})`}
-                  size="small" shape="circular" appearance="secondary" />
-                <ToolTipButton desc={t('Remove Attachment')}
-                  icon={<Dismiss16Regular />}
-                  size="small" shape="circular" appearance="subtle"
-                  onClick={() => {
-                    commonStore.setCurrentTempAttachment(null);
-                  }} />
-              </div>
-            }
+                    desc={`${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();
   }, []);
 
@@ -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

@@ -82,8 +82,10 @@ const CompositionPanel: FC = observer(() => {
   };
 
   useEffect(() => {
-    if (inputRef.current)
+    if (inputRef.current) {
       inputRef.current.style.height = '100%';
+      inputRef.current.style.maxHeight = '100%';
+    }
     scrollToBottom();
 
     if (playerRef.current && visualizerRef.current) {
@@ -167,6 +169,8 @@ 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+$/, ''));

frontend/src/pages/Configs.tsx

@@ -150,7 +150,7 @@ 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({
@@ -328,7 +328,7 @@ const Configs: FC = observer(() => {
                 } />
                 {
                   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"
                         value={selectedConfig.modelParameters.precision}
@@ -340,8 +340,10 @@ const Configs: FC = observer(() => {
                             });
                           }
                         }}>
-                        <Option>fp16</Option>
+                        {selectedConfig.modelParameters.device !== 'CPU' && selectedConfig.modelParameters.device !== 'MPS' &&
+                          <Option>fp16</Option>}
                         <Option>int8</Option>
+                        {selectedConfig.modelParameters.device === 'WebGPU' && <Option>nf4</Option>}
                         {selectedConfig.modelParameters.device !== 'WebGPU' && <Option>fp32</Option>}
                       </Dropdown>
                     } />

frontend/src/pages/Models.tsx

@@ -1,5 +1,6 @@
 import React, { FC } from 'react';
 import {
+  Checkbox,
   createTableColumn,
   DataGrid,
   DataGridBody,
@@ -19,7 +20,7 @@ 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 +140,7 @@ 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!);
@@ -160,10 +161,21 @@ 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.')}

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/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>
               )

frontend/src/pages/Train.tsx

@@ -129,11 +129,12 @@ 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',
+  '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) => {

frontend/src/stores/commonStore.ts

@@ -7,7 +7,7 @@ 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 { ModelConfig } from '../types/configs';
@@ -51,6 +51,7 @@ class CommonStore {
   monitorData: MonitorData | null = null;
   depComplete: boolean = false;
   platform: Platform = 'windows';
+  lastModelName: string = '';
   // presets manager
   editingPreset: Preset | null = null;
   presets: Preset[] = [];
@@ -64,6 +65,14 @@ class CommonStore {
   attachmentUploading: boolean = false;
   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;
@@ -133,6 +142,7 @@ class CommonStore {
     autoUpdatesCheck: true,
     giteeUpdatesSource: getUserLanguage() === 'zh',
     cnMirror: getUserLanguage() === 'zh',
+    useHfMirror: false,
     host: '127.0.0.1',
     dpiScaling: 100,
     customModelsPath: './models',
@@ -232,6 +242,10 @@ class CommonStore {
     this.about = value;
   };
 
+  setLastModelName(value: string) {
+    this.lastModelName = value;
+  }
+
   setDepComplete = (value: boolean, inSaveCache: boolean = true) => {
     this.depComplete = value;
     if (inSaveCache)
@@ -358,6 +372,14 @@ 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;
+  }
 }
 
 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/configs.ts

@@ -7,7 +7,7 @@ export type ApiParameters = {
   frequencyPenalty: number;
 }
 export type Device = 'CPU' | 'CUDA' | 'CUDA-Beta' | 'WebGPU' | 'MPS' | 'Custom';
-export type Precision = 'fp16' | 'int8' | 'fp32';
+export type Precision = 'fp16' | 'int8' | 'fp32' | 'nf4';
 export type ModelParameters = {
   // different models can not have the same name
   modelName: string;

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

frontend/src/utils/index.tsx

@@ -178,14 +178,14 @@ export const getStrategy = (modelConfig: ModelConfig | undefined = undefined) =>
       strategy += params.precision === 'int8' ? 'fp32i8' : 'fp32';
       break;
     case 'WebGPU':
-      strategy += params.precision === 'int8' ? 'fp16i8' : 'fp16';
+      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 +193,7 @@ 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 || '';
@@ -290,12 +290,15 @@ export function bytesToReadable(size: number) {
 }
 
 export function getServerRoot(defaultLocalPort: number) {
+  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) {
@@ -471,6 +474,12 @@ 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 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)))

manifest.json

@@ -1,5 +1,5 @@
 {
-  "version": "1.5.1",
+  "version": "1.5.2",
   "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）。"