Compare commits

...

6 Commits

Author SHA1 Message Date
josc146
3d5c3dcd31 release v1.5.2 2023-11-09 22:11:05 +08:00
josc146
0a4876a564 improve user guide 2023-11-09 22:07:01 +08:00
josc146
4f0558ae34 add client upgrade progress 2023-11-09 21:38:02 +08:00
josc146
f03c9cf25f improve mobile view 2023-11-09 12:21:01 +08:00
josc146
07797537d1 add RWKV-Runner WebUI to Server-Deploy-Examples 2023-11-09 00:21:02 +08:00
github-actions[bot]
0c3a50cb07 release v1.5.1 2023-11-08 15:41:53 +00:00
16 changed files with 218 additions and 115 deletions

View File

@@ -1,22 +1,10 @@
## Changes
### Features
### Improvements
- add webUI for easier service sharing (enable it in Configs page or --webui command line parameter, compile it
with `make
build-web`)
- add deployment mode. If `/switch-model` with `deploy: true`, will disable /switch-model, /exit and other dangerous
APIs (state cache APIs, part of midi APIs)
### Chores
- print error.txt to console when script fails
- api url getter
### Fixes
- set deepspeed to 0.11.2 to avoid finetune error
- fix `/docs` default api params (Pydantic v2)
- improve mobile webview
- add client upgrade progress
- improve user guide
## Install

View File

@@ -4,12 +4,14 @@ import (
"bufio"
"context"
"errors"
"io"
"net/http"
"os"
"os/exec"
"path/filepath"
"runtime"
"syscall"
"time"
"github.com/fsnotify/fsnotify"
"github.com/minio/selfupdate"
@@ -118,13 +120,50 @@ func (a *App) monitorHardware() {
monitor.Start()
}
type ProgressReader struct {
reader io.Reader
total int64
err error
}
func (pr *ProgressReader) Read(p []byte) (n int, err error) {
n, err = pr.reader.Read(p)
pr.err = err
pr.total += int64(n)
return
}
func (a *App) UpdateApp(url string) (broken bool, err error) {
resp, err := http.Get(url)
if err != nil {
return false, err
}
defer resp.Body.Close()
err = selfupdate.Apply(resp.Body, selfupdate.Options{})
pr := &ProgressReader{reader: resp.Body}
ticker := time.NewTicker(250 * time.Millisecond)
defer ticker.Stop()
go func() {
for {
<-ticker.C
wruntime.EventsEmit(a.ctx, "updateApp", &DownloadStatus{
Name: filepath.Base(url),
Path: "",
Url: url,
Transferred: pr.total,
Size: resp.ContentLength,
Speed: 0,
Progress: 100 * (float64(pr.total) / float64(resp.ContentLength)),
Downloading: pr.err == nil && pr.total < resp.ContentLength,
Done: pr.total == resp.ContentLength,
})
if pr.err != nil || pr.total == resp.ContentLength {
break
}
}
}()
err = selfupdate.Apply(pr, selfupdate.Options{})
if err != nil {
if rerr := selfupdate.RollbackError(err); rerr != nil {
return true, rerr

View File

@@ -0,0 +1,18 @@
: install git python3.10 npm by yourself
: change model and strategy according to your hardware
git clone https://github.com/josStorer/RWKV-Runner --depth=1
python -m pip install torch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 --index-url https://download.pytorch.org/whl/cu117
python -m pip install -r RWKV-Runner/backend-python/requirements.txt
cd RWKV-Runner/frontend
call npm ci
call npm run build
cd ..
start python ./backend-python/main.py --webui
start "C:\Program Files (x86)\Microsoft\Edge\Application\msedge.exe" "http://127.0.0.1:8000"
powershell -Command "(Test-Path ./models) -or (mkdir models)"
powershell -Command "Import-Module BitsTransfer"
powershell -Command "(Test-Path ./models/RWKV-4-World-1.5B-v1-fixed-20230612-ctx4096.pth) -or (Start-BitsTransfer https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-1.5B-v1-fixed-20230612-ctx4096.pth ./models/RWKV-4-World-1.5B-v1-fixed-20230612-ctx4096.pth)"
powershell -Command "Invoke-WebRequest http://127.0.0.1:8000/switch-model -Method POST -ContentType 'application/json' -Body '{\"model\":\"./models/RWKV-4-World-1.5B-v1-fixed-20230612-ctx4096.pth\",\"strategy\":\"cuda fp32 *20+\",\"deploy\":\"true\"}'"

View File

@@ -0,0 +1,21 @@
# install git python3.10 npm by yourself
# change model and strategy according to your hardware
sudo apt install python3-dev
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
cd RWKV-Runner/frontend
npm ci
npm run build
cd ..
python3 ./backend-python/main.py --webui > log.txt &
if [ ! -d models ]; then
mkdir models
fi
wget -N https://huggingface.co/BlinkDL/rwkv-4-world/resolve/main/RWKV-4-World-0.1B-v1-20230520-ctx4096.pth -P models/
curl http://127.0.0.1:8000/switch-model -X POST -H "Content-Type: application/json" -d '{"model":"./models/RWKV-4-World-0.1B-v1-20230520-ctx4096.pth","strategy":"cpu fp32","deploy":"true"}'

View File

@@ -51,10 +51,10 @@ const App: FC = observer(() => {
useEffect(() => setPath(location.pathname), [location]);
return (
<FluentProvider className="h-screen"
<FluentProvider
theme={commonStore.settings.darkMode ? webDarkTheme : webLightTheme}
data-theme={commonStore.settings.darkMode ? 'dark' : 'light'}>
<div className="flex h-full">
<div className="flex h-screen">
<div className="flex flex-col w-16 sm:w-48 p-2 justify-between">
<TabList
size="large"

View File

@@ -13,7 +13,7 @@
"Working": "動作中",
"Stop": "停止",
"Enable High Precision For Last Layer": "最後の層で高精度を有効にする",
"Stored Layers": "メモリ層読み込み",
"Stored Layers": "保存されるレイヤー",
"Precision": "精度",
"Device": "デバイス",
"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.": "これらの設定でモデルを変換します。変換されたモデルを使用すると、読み込み速度が大幅に向上しますが、変換したモデルのパラメータを変更することはできません。",
@@ -267,5 +267,6 @@
"Hello, what can I do for you?": "こんにちは、何かお手伝いできますか?",
"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"
"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"
}

View File

@@ -267,5 +267,6 @@
"Hello, what can I do for you?": "你好,有什么要我帮忙的吗?",
"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"
"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"
}

View File

@@ -213,6 +213,12 @@ export const RunButton: FC<{ onClickRun?: MouseEventHandler, iconMode?: boolean
const buttonFn = () => {
navigate({ pathname: '/' + buttonName.toLowerCase() });
};
if ((modelConfig.modelParameters.device === 'CUDA' || modelConfig.modelParameters.device === 'CUDA-Beta') &&
modelConfig.modelParameters.storedLayers < modelConfig.modelParameters.maxStoredLayers &&
commonStore.monitorData && commonStore.monitorData.totalVram !== 0 &&
(commonStore.monitorData.usedVram / commonStore.monitorData.totalVram) < 0.85)
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) {
toast(t('Loading Model'), { type: 'info' });

View File

@@ -19,11 +19,13 @@ import { defaultCompositionPrompt } from './defaultConfigs';
import { FileExists, OpenFileFolder, OpenSaveFileDialogBytes } from '../../wailsjs/go/backend_golang/App';
import { getServerRoot, toastWithButton } from '../utils';
import { CompositionParams } from '../types/composition';
import { useMediaQuery } from 'usehooks-ts';
let compositionSseController: AbortController | null = null;
const CompositionPanel: FC = observer(() => {
const { t } = useTranslation();
const mq = useMediaQuery('(min-width: 640px)');
const inputRef = useRef<HTMLTextAreaElement>(null);
const port = commonStore.getCurrentModelConfig().apiParameters.apiPort;
const visualizerRef = useRef<VisualizerElement>(null);
@@ -204,62 +206,64 @@ const CompositionPanel: FC = observer(() => {
setPrompt(e.target.value);
}}
/>
<div className="flex flex-col gap-1 max-h-48 sm:max-w-sm sm:max-h-full 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={4100}
step={100}
input
onChange={(e, data) => {
setParams({
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) => {
setParams({
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) => {
setParams({
topP: data.value
});
}} />
} />
<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;
<div className="flex flex-col gap-1 max-h-48 sm:max-w-sm sm:max-h-full">
<div className="flex flex-col gap-1 grow 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={4100}
step={100}
input
onChange={(e, data) => {
setParams({
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) => {
setParams({
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) => {
setParams({
topP: data.value
});
}} />
} />
<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;
}
}
}
setParams({
useLocalSoundFont: data.checked as boolean
});
setSoundFont();
}} />
<Checkbox className="select-none"
size="large" label={t('Auto Play At The End')} checked={params.autoPlay} onChange={(_, data) => {
setParams({
useLocalSoundFont: data.checked as boolean
autoPlay: data.checked as boolean
});
setSoundFont();
}} />
<Checkbox className="select-none"
size="large" label={t('Auto Play At The End')} checked={params.autoPlay} onChange={(_, data) => {
setParams({
autoPlay: data.checked as boolean
});
}} />
</div>
<div className="flex justify-between gap-2">
<ToolTipButton desc={t('Regenerate')} icon={<ArrowSync20Regular />} onClick={() => {
compositionSseController?.abort();
@@ -298,7 +302,7 @@ const CompositionPanel: FC = observer(() => {
ref={playerRef}
style={{ width: '100%' }}
/>
<Button icon={<Save28Regular />}
<Button icon={<Save28Regular />} size={mq ? 'large' : 'medium'} appearance={mq ? 'secondary' : 'subtle'}
onClick={() => {
if (params.midi) {
OpenSaveFileDialogBytes('*.mid', 'music.mid', Array.from(new Uint8Array(params.midi))).then((path) => {
@@ -314,7 +318,7 @@ const CompositionPanel: FC = observer(() => {
}
}}
>
{t('Save')}
{mq ? t('Save') : ''}
</Button>
</div>
</div>

View File

@@ -89,7 +89,7 @@ const Home: FC = observer(() => {
return commonStore.platform === 'web' ?
(
<div className="flex flex-col gap-2 h-full">
<div className="flex flex-col gap-2 h-full overflow-x-hidden overflow-y-auto">
<img className="rounded-xl select-none object-cover grow"
style={{ maxHeight: '40%' }} src={banner} />
<div className="grow"></div>

View File

@@ -1,4 +1,4 @@
import commonStore, { Platform } from './stores/commonStore';
import commonStore, { MonitorData, Platform } from './stores/commonStore';
import { GetPlatform, ListDirFiles, ReadJson } from '../wailsjs/go/backend_golang/App';
import { Cache, checkUpdate, downloadProgramFiles, LocalConfig, refreshLocalModels, refreshModels } from './utils';
import { getStatus } from './apis';
@@ -125,19 +125,12 @@ async function initLocalModelsNotify() {
});
}
type monitorData = {
usedMemory: number;
totalMemory: number;
gpuUsage: number;
gpuPower: number;
usedVram: number;
totalVram: number;
}
async function initHardwareMonitor() {
EventsOn('monitor', (data: string) => {
const results: monitorData = JSON.parse(data);
if (results)
const results: MonitorData = JSON.parse(data);
if (results) {
commonStore.setMonitorData(results);
WindowSetTitle(`RWKV-Runner (${t('RAM')}: ${results.usedMemory.toFixed(1)}/${results.totalMemory.toFixed(1)} GB, ${t('VRAM')}: ${(results.usedVram / 1024).toFixed(1)}/${(results.totalVram / 1024).toFixed(1)} GB, ${t('GPU Usage')}: ${results.gpuUsage}%)`);
}
});
}

View File

@@ -7,7 +7,7 @@ import { defaultCompositionPrompt, defaultModelConfigs, defaultModelConfigsMac }
import { ChartData } from 'chart.js';
import { Preset } from '../types/presets';
import { AboutContent } from '../types/about';
import { Conversation } from '../types/chat';
import { Attachment, Conversation } from '../types/chat';
import { CompletionPreset } from '../types/completion';
import { CompositionParams } from '../types/composition';
import { ModelConfig } from '../types/configs';
@@ -30,10 +30,13 @@ export type Status = {
device_name: string;
}
export type Attachment = {
name: string;
size: number;
content: string;
export type MonitorData = {
usedMemory: number;
totalMemory: number;
gpuUsage: number;
gpuPower: number;
usedVram: number;
totalVram: number;
}
export type Platform = 'windows' | 'darwin' | 'linux' | 'web';
@@ -45,6 +48,7 @@ class CommonStore {
pid: 0,
device_name: 'CPU'
};
monitorData: MonitorData | null = null;
depComplete: boolean = false;
platform: Platform = 'windows';
// presets manager
@@ -254,6 +258,10 @@ class CommonStore {
this.completionGenerating = value;
}
setMonitorData(value: MonitorData) {
this.monitorData = value;
}
setPlatform(value: Platform) {
this.platform = value;
}

View File

@@ -14,7 +14,6 @@
body {
margin: 0;
overflow: hidden;
width: 100%;
height: 100%;
}

View File

@@ -26,4 +26,9 @@ export type Role = 'assistant' | 'user' | 'system';
export type ConversationMessage = {
role: Role;
content: string;
}
export type Attachment = {
name: string;
size: number;
content: string;
}

View File

@@ -15,7 +15,7 @@ import { toast } from 'react-toastify';
import { t } from 'i18next';
import { ToastOptions } from 'react-toastify/dist/types';
import { Button } from '@fluentui/react-components';
import { BrowserOpenURL, WindowShow } from '../../wailsjs/runtime';
import { BrowserOpenURL, EventsOff, EventsOn, WindowShow } from '../../wailsjs/runtime';
import { NavigateFunction } from 'react-router';
import { ModelConfig, ModelParameters } from '../types/configs';
import { DownloadStatus } from '../types/downloads';
@@ -333,27 +333,47 @@ export async function checkUpdate(notifyEvenLatest: boolean = false) {
`https://gitee.com/josc146/RWKV-Runner/releases/download/${versionTag}/${asset.name}`;
toastWithButton(t('New Version Available') + ': ' + versionTag, t('Update'), () => {
DeleteFile('cache.json');
toast(t('Downloading update, please wait. If it is not completed, please manually download the program from GitHub and replace the original program.'), {
type: 'info',
position: 'bottom-left',
autoClose: false
});
setTimeout(() => {
UpdateApp(updateUrl).then(() => {
toast(t('Update completed, please restart the program.'), {
type: 'success',
position: 'bottom-left',
autoClose: false
}
);
}).catch((e) => {
toast(t('Update Error') + ' - ' + (e.message || e), {
type: 'error',
const progressId = 'update_app';
const progressEvent = 'updateApp';
const updateProgress = (ds: DownloadStatus | null) => {
const content =
t('Downloading update, please wait. If it is not completed, please manually download the program from GitHub and replace the original program.')
+ (ds ? ` (${ds.progress.toFixed(2)}% ${bytesToReadable(ds.transferred)}/${bytesToReadable(ds.size)})` : '');
const options: ToastOptions = {
type: 'info',
position: 'bottom-left',
autoClose: false,
toastId: progressId,
hideProgressBar: false,
progress: ds ? ds.progress / 100 : 0
};
if (toast.isActive(progressId))
toast.update(progressId, {
render: content,
...options
});
else
toast(content, options);
};
updateProgress(null);
EventsOn(progressEvent, updateProgress);
UpdateApp(updateUrl).then(() => {
toast(t('Update completed, please restart the program.'), {
type: 'success',
position: 'bottom-left',
autoClose: false
});
}
);
}).catch((e) => {
toast(t('Update Error') + ' - ' + (e.message || e), {
type: 'error',
position: 'bottom-left',
autoClose: false
});
}, 500);
}).finally(() => {
toast.dismiss(progressId);
EventsOff(progressEvent);
});
}, {
autoClose: false,
position: 'bottom-left'

View File

@@ -1,5 +1,5 @@
{
"version": "1.5.0",
"version": "1.5.1",
"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)。"