release v1.2.8

CURRENT_CHANGE.md

@@ -1,8 +1,7 @@
 ## Changes
 
-- add DPI Scaling setting
-- allow custom api url, key, model
-- add chat and completion error messages
+- exact avoidOverflow
+- adjust MoreUtilsButton
 
 ## Install
 

README_ZH.md

@@ -46,7 +46,7 @@ API兼容的接口，这意味着一切ChatGPT客户端都是RWKV客户端。
 
 </div>
 
-#### 注意 目前RWKV中文模型质量一般，推荐使用英文模型或World(100+ 语言)体验实际RWKV能力
+#### 注意 目前RWKV中文模型质量一般，推荐使用英文模型或World(全球语言)体验实际RWKV能力
 
 #### 预设配置已经开启自定义CUDA算子加速，速度更快，且显存消耗更少。如果你遇到可能的兼容性问题，前往配置页面，关闭`使用自定义CUDA算子加速`
 

backend-golang/rwkv.go

@@ -4,7 +4,6 @@ import (
 	"errors"
 	"os"
 	"os/exec"
-	"path/filepath"
 	"runtime"
 	"strconv"
 	"strings"
@@ -52,11 +51,7 @@ func (a *App) InstallPyDep(python string, cnMirror bool) (string, error) {
 	if python == "" {
 		python, err = GetPython()
 		if runtime.GOOS == "windows" {
-			python, err = filepath.Abs(python)
-			if err != nil {
-				return "", err
-			}
-			python = `"` + python + `"`
+			python = `"%CD%/` + python + `"`
 		}
 	}
 	if err != nil {

backend-python/utils/rwkv.py

@@ -325,7 +325,6 @@ The following is a coherent verbose detailed conversation between a girl named {
                 response += delta
                 if stop is not None:
                     if stop in response:
-                        response = response.split(stop)[0]
                         try:
                             state_cache.add_state(
                                 state_cache.AddStateBody(
@@ -337,6 +336,7 @@ The following is a coherent verbose detailed conversation between a girl named {
                             )
                         except HTTPException:
                             pass
+                        response = response.split(stop)[0]
                         yield response, "", prompt_token_len, completion_token_len
                         break
                 out_last = begin + i + 1

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

@@ -154,5 +154,8 @@
   "Restart": "重启",
   "API Chat Model Name": "API聊天模型名",
   "API Completion Model Name": "API补全模型名",
-  "Localhost": "本地"
+  "Localhost": "本地",
+  "Retry": "重试",
+  "Delete": "删除",
+  "Edit": "编辑"
 }

frontend/src/components/CopyButton.tsx

@@ -4,7 +4,7 @@ import { useTranslation } from 'react-i18next';
 import { ClipboardSetText } from '../../wailsjs/runtime';
 import { ToolTipButton } from './ToolTipButton';
 
-export const CopyButton: FC<{ content: string }> = ({ content }) => {
+export const CopyButton: FC<{ content: string, showDelay?: number, }> = ({ content, showDelay = 0 }) => {
   const { t } = useTranslation();
   const [copied, setCopied] = useState(false);
 
@@ -19,7 +19,8 @@ export const CopyButton: FC<{ content: string }> = ({ content }) => {
   };
 
   return (
-    <ToolTipButton desc={t('Copy')} size="small" appearance="subtle" icon={copied ? <CheckIcon /> : <CopyIcon />}
+    <ToolTipButton desc={t('Copy')} size="small" appearance="subtle" showDelay={showDelay}
+      icon={copied ? <CheckIcon /> : <CopyIcon />}
       onClick={onClick} />
   );
 };

frontend/src/components/ReadButton.tsx

@@ -7,13 +7,28 @@ import { observer } from 'mobx-react-lite';
 
 const synth = window.speechSynthesis;
 
-export const ReadButton: FC<{ content: string }> = observer(({ content }) => {
+export const ReadButton: FC<{
+  content: string,
+  inSpeaking?: boolean,
+  showDelay?: number,
+  setSpeakingOuter?: (speaking: boolean) => void
+}> = observer(({
+  content,
+  inSpeaking = false,
+  showDelay = 0,
+  setSpeakingOuter
+}) => {
   const { t } = useTranslation();
-  const [speaking, setSpeaking] = useState(false);
+  const [speaking, setSpeaking] = useState(inSpeaking);
   let lang: string = commonStore.settings.language;
   if (lang === 'dev')
     lang = 'en';
 
+  const setSpeakingInner = (speaking: boolean) => {
+    setSpeakingOuter?.(speaking);
+    setSpeaking(speaking);
+  };
+
   const startSpeak = () => {
     synth.cancel();
 
@@ -31,22 +46,22 @@ export const ReadButton: FC<{ content: string }> = observer(({ content }) => {
     Object.assign(utterance, {
       rate: 1,
       volume: 1,
-      onend: () => setSpeaking(false),
-      onerror: () => setSpeaking(false),
+      onend: () => setSpeakingInner(false),
+      onerror: () => setSpeakingInner(false),
       voice: voice
     });
 
     synth.speak(utterance);
-    setSpeaking(true);
+    setSpeakingInner(true);
   };
 
   const stopSpeak = () => {
     synth.cancel();
-    setSpeaking(false);
+    setSpeakingInner(false);
   };
 
   return (
-    <ToolTipButton desc={t('Read Aloud')} size="small" appearance="subtle"
+    <ToolTipButton desc={t('Read Aloud')} size="small" appearance="subtle" showDelay={showDelay}
       icon={speaking ? <MuteIcon /> : <UnmuteIcon />}
       onClick={speaking ? stopSpeak : startSpeak} />
   );

frontend/src/components/RunButton.tsx

@@ -167,9 +167,10 @@ export const RunButton: FC<{ onClickRun?: MouseEventHandler, iconMode?: boolean
               frequency_penalty: modelConfig.apiParameters.frequencyPenalty
             });
 
+            const strategy = getStrategy(modelConfig);
             let customCudaFile = '';
             if ((modelConfig.modelParameters.device === 'CUDA' || modelConfig.modelParameters.device === 'Custom')
-              && modelConfig.modelParameters.useCustomCuda) {
+              && modelConfig.modelParameters.useCustomCuda && !strategy.includes('fp32')) {
               if (commonStore.platform === 'windows') {
                 customCudaFile = getSupportedCustomCudaFile();
                 if (customCudaFile) {
@@ -194,7 +195,7 @@ export const RunButton: FC<{ onClickRun?: MouseEventHandler, iconMode?: boolean
 
             switchModel({
               model: modelPath,
-              strategy: getStrategy(modelConfig),
+              strategy: strategy,
               customCuda: customCudaFile !== ''
             }).then(async (r) => {
               if (r.ok) {

frontend/src/components/ToolTipButton.tsx

@@ -11,6 +11,7 @@ export const ToolTipButton: FC<{
   appearance?: 'secondary' | 'primary' | 'outline' | 'subtle' | 'transparent';
   disabled?: boolean,
   onClick?: MouseEventHandler
+  showDelay?: number,
 }> = ({
   text,
   desc,
@@ -20,10 +21,11 @@ export const ToolTipButton: FC<{
   shape,
   appearance,
   disabled,
-  onClick
+  onClick,
+  showDelay = 0
 }) => {
   return (
-    <Tooltip content={desc} showDelay={0} hideDelay={0} relationship="label">
+    <Tooltip content={desc} showDelay={showDelay} hideDelay={0} relationship="label">
       <Button className={className} disabled={disabled} icon={icon} onClick={onClick} size={size} shape={shape}
         appearance={appearance}>{text}</Button>
     </Tooltip>

frontend/src/pages/Chat.tsx

@@ -1,12 +1,13 @@
-import React, { FC, useEffect, useRef } from 'react';
+import React, { FC, useCallback, useEffect, useRef, useState } from 'react';
 import { useTranslation } from 'react-i18next';
-import { Avatar, PresenceBadge, Textarea } from '@fluentui/react-components';
+import { Avatar, Button, Menu, MenuPopover, MenuTrigger, PresenceBadge, Textarea } from '@fluentui/react-components';
 import commonStore, { ModelStatus } from '../stores/commonStore';
 import { observer } from 'mobx-react-lite';
 import { v4 as uuid } from 'uuid';
 import classnames from 'classnames';
 import { fetchEventSource } from '@microsoft/fetch-event-source';
-import { ConversationPair, getConversationPairs, Record } from '../utils/get-conversation-pairs';
+import { KebabHorizontalIcon, PencilIcon, SyncIcon, TrashIcon } from '@primer/octicons-react';
+import { ConversationPair } from '../utils/get-conversation-pairs';
 import logo from '../assets/images/logo.jpg';
 import MarkdownRender from '../components/MarkdownRender';
 import { ToolTipButton } from '../components/ToolTipButton';
@@ -22,6 +23,8 @@ import { toastWithButton } from '../utils';
 export const userName = 'M E';
 export const botName = 'A I';
 
+export const welcomeUuid = 'welcome';
+
 export enum MessageType {
   Normal,
   Error
@@ -48,6 +51,126 @@ export type Conversation = {
 
 let chatSseController: AbortController | null = null;
 
+const MoreUtilsButton: FC<{ uuid: string, setEditing: (editing: boolean) => void }> = observer(({
+  uuid,
+  setEditing
+}) => {
+  const { t } = useTranslation();
+  const [speaking, setSpeaking] = useState(false);
+
+  const messageItem = commonStore.conversation[uuid];
+
+  return <Menu>
+    <MenuTrigger disableButtonEnhancement>
+      <Button icon={<KebabHorizontalIcon />} size="small" appearance="subtle" />
+    </MenuTrigger>
+    <MenuPopover style={{ minWidth: 0 }}>
+      <CopyButton content={messageItem.content} showDelay={500} />
+      <ReadButton content={messageItem.content} inSpeaking={speaking} showDelay={500} setSpeakingOuter={setSpeaking} />
+      <ToolTipButton desc={t('Edit')} icon={<PencilIcon />} showDelay={500} size="small" appearance="subtle"
+        onClick={() => {
+          setEditing(true);
+        }} />
+      <ToolTipButton desc={t('Delete')} icon={<TrashIcon />} showDelay={500} size="small" appearance="subtle"
+        onClick={() => {
+          commonStore.conversationOrder.splice(commonStore.conversationOrder.indexOf(uuid), 1);
+          delete commonStore.conversation[uuid];
+        }} />
+    </MenuPopover>
+  </Menu>;
+});
+
+const ChatMessageItem: FC<{
+  uuid: string, onSubmit: (message: string | null, answerId: string | null,
+    startUuid: string | null, endUuid: string | null, includeEndUuid: boolean) => void
+}> = observer(({ uuid, onSubmit }) => {
+  const { t } = useTranslation();
+  const [editing, setEditing] = useState(false);
+  const textareaRef = useRef<HTMLTextAreaElement>(null);
+  const messageItem = commonStore.conversation[uuid];
+
+  console.log(uuid);
+
+  const setEditingInner = (editing: boolean) => {
+    setEditing(editing);
+    if (editing) {
+      setTimeout(() => {
+        const textarea = textareaRef.current;
+        if (textarea) {
+          textarea.focus();
+          textarea.selectionStart = textarea.value.length;
+          textarea.selectionEnd = textarea.value.length;
+          textarea.style.height = textarea.scrollHeight + 'px';
+        }
+      });
+    }
+  };
+
+  return <div
+    className={classnames(
+      'flex gap-2 mb-2 overflow-hidden',
+      messageItem.side === 'left' ? 'flex-row' : 'flex-row-reverse'
+    )}
+    onMouseEnter={() => {
+      const utils = document.getElementById('utils-' + uuid);
+      if (utils) utils.classList.remove('invisible');
+    }}
+    onMouseLeave={() => {
+      const utils = document.getElementById('utils-' + uuid);
+      if (utils) utils.classList.add('invisible');
+    }}
+  >
+    <Avatar
+      color={messageItem.color}
+      name={messageItem.sender}
+      image={messageItem.avatarImg ? { src: messageItem.avatarImg } : undefined}
+    />
+    <div
+      className={classnames(
+        'flex p-2 rounded-lg overflow-hidden',
+        editing ? 'grow' : '',
+        messageItem.side === 'left' ? 'bg-gray-200' : 'bg-blue-500',
+        messageItem.side === 'left' ? 'text-gray-600' : 'text-white'
+      )}
+    >
+      {!editing ?
+        <MarkdownRender>{messageItem.content}</MarkdownRender> :
+        <Textarea ref={textareaRef}
+          className="grow"
+          style={{ minWidth: 0 }}
+          value={messageItem.content}
+          onChange={(e) => {
+            messageItem.content = e.target.value;
+          }}
+          onBlur={() => {
+            setEditingInner(false);
+          }} />}
+    </div>
+    <div className="flex flex-col gap-1 items-start">
+      <div className="grow" />
+      {(messageItem.type === MessageType.Error || !messageItem.done) &&
+        <PresenceBadge size="extra-small" status={
+          messageItem.type === MessageType.Error ? 'busy' : 'away'
+        } />
+      }
+      <div className="flex invisible" id={'utils-' + uuid}>
+        {
+          messageItem.sender === botName && uuid !== welcomeUuid &&
+          <ToolTipButton desc={t('Retry')} size="small" appearance="subtle"
+            icon={<SyncIcon />} onClick={() => {
+            onSubmit(null, uuid, null, uuid, false);
+          }} />
+        }
+        <ToolTipButton desc={t('Edit')} icon={<PencilIcon />} size="small" appearance="subtle"
+          onClick={() => {
+            setEditingInner(true);
+          }} />
+        <MoreUtilsButton uuid={uuid} setEditing={setEditingInner} />
+      </div>
+    </div>
+  </div>;
+});
+
 const ChatPanel: FC = observer(() => {
   const { t } = useTranslation();
   const bodyRef = useRef<HTMLDivElement>(null);
@@ -71,9 +194,9 @@ const ChatPanel: FC = observer(() => {
 
   useEffect(() => {
     if (commonStore.conversationOrder.length === 0) {
-      commonStore.setConversationOrder(['welcome']);
+      commonStore.setConversationOrder([welcomeUuid]);
       commonStore.setConversation({
-        'welcome': {
+        [welcomeUuid]: {
           sender: botName,
           type: MessageType.Normal,
           color: 'colorful',
@@ -106,38 +229,48 @@ const ChatPanel: FC = observer(() => {
     }
   };
 
-  const onSubmit = (message: string) => {
-    const newId = uuid();
-    commonStore.conversation[newId] = {
-      sender: userName,
-      type: MessageType.Normal,
-      color: 'brand',
-      time: new Date().toISOString(),
-      content: message,
-      side: 'right',
-      done: true
-    };
-    commonStore.setConversation(commonStore.conversation);
-    commonStore.conversationOrder.push(newId);
-    commonStore.setConversationOrder(commonStore.conversationOrder);
+  // if message is not null, create a user message;
+  // if answerId is not null, override the answer with new response;
+  // if startUuid is null, start generating api body messages from first message;
+  // if endUuid is null, generate api body messages until last message;
+  const onSubmit = useCallback((message: string | null = null, answerId: string | null = null,
+    startUuid: string | null = null, endUuid: string | null = null, includeEndUuid: boolean = false) => {
+    if (message) {
+      const newId = uuid();
+      commonStore.conversation[newId] = {
+        sender: userName,
+        type: MessageType.Normal,
+        color: 'brand',
+        time: new Date().toISOString(),
+        content: message,
+        side: 'right',
+        done: true
+      };
+      commonStore.setConversation(commonStore.conversation);
+      commonStore.conversationOrder.push(newId);
+      commonStore.setConversationOrder(commonStore.conversationOrder);
+    }
 
-    const records: Record[] = [];
-    commonStore.conversationOrder.forEach((uuid, index) => {
+    let startIndex = startUuid ? commonStore.conversationOrder.indexOf(startUuid) : 0;
+    let endIndex = endUuid ? (commonStore.conversationOrder.indexOf(endUuid) + (includeEndUuid ? 1 : 0)) : commonStore.conversationOrder.length;
+    let targetRange = commonStore.conversationOrder.slice(startIndex, endIndex);
+
+    const messages: ConversationPair[] = [];
+    targetRange.forEach((uuid, index) => {
+      if (uuid === welcomeUuid)
+        return;
       const messageItem = commonStore.conversation[uuid];
-      if (messageItem.done && messageItem.type === MessageType.Normal && messageItem.sender === botName) {
-        if (index > 0) {
-          const questionId = commonStore.conversationOrder[index - 1];
-          const question = commonStore.conversation[questionId];
-          if (question.done && question.type === MessageType.Normal && question.sender === userName) {
-            records.push({ question: question.content, answer: messageItem.content });
-          }
-        }
+      if (messageItem.done && messageItem.type === MessageType.Normal && messageItem.sender === userName) {
+        messages.push({ role: 'user', content: messageItem.content });
+      } else if (messageItem.done && messageItem.type === MessageType.Normal && messageItem.sender === botName) {
+        messages.push({ role: 'assistant', content: messageItem.content });
       }
     });
-    const messages = getConversationPairs(records, false);
-    (messages as ConversationPair[]).push({ role: 'user', content: message });
 
-    const answerId = uuid();
+    if (answerId === null) {
+      answerId = uuid();
+      commonStore.conversationOrder.push(answerId);
+    }
     commonStore.conversation[answerId] = {
       sender: botName,
       type: MessageType.Normal,
@@ -149,7 +282,6 @@ const ChatPanel: FC = observer(() => {
       done: false
     };
     commonStore.setConversation(commonStore.conversation);
-    commonStore.conversationOrder.push(answerId);
     commonStore.setConversationOrder(commonStore.conversationOrder);
     setTimeout(scrollToBottom);
     let answer = '';
@@ -171,11 +303,10 @@ const ChatPanel: FC = observer(() => {
         }),
         signal: chatSseController?.signal,
         onmessage(e) {
-          console.log('sse message', e);
           scrollToBottom();
           if (e.data === '[DONE]') {
-            commonStore.conversation[answerId].done = true;
-            commonStore.conversation[answerId].content = commonStore.conversation[answerId].content.trim();
+            commonStore.conversation[answerId!].done = true;
+            commonStore.conversation[answerId!].content = commonStore.conversation[answerId!].content.trim();
             commonStore.setConversation(commonStore.conversation);
             commonStore.setConversationOrder([...commonStore.conversationOrder]);
             return;
@@ -189,14 +320,14 @@ const ChatPanel: FC = observer(() => {
           }
           if (data.choices && Array.isArray(data.choices) && data.choices.length > 0) {
             answer += data.choices[0]?.delta?.content || '';
-            commonStore.conversation[answerId].content = answer;
+            commonStore.conversation[answerId!].content = answer;
             commonStore.setConversation(commonStore.conversation);
             commonStore.setConversationOrder([...commonStore.conversationOrder]);
           }
         },
         async onopen(response) {
           if (response.status !== 200) {
-            commonStore.conversation[answerId].content += '\n[ERROR]\n```\n' + response.statusText + '\n' + (await response.text()) + '\n```';
+            commonStore.conversation[answerId!].content += '\n[ERROR]\n```\n' + response.statusText + '\n' + (await response.text()) + '\n```';
             commonStore.setConversation(commonStore.conversation);
             commonStore.setConversationOrder([...commonStore.conversationOrder]);
             setTimeout(scrollToBottom);
@@ -206,67 +337,25 @@ const ChatPanel: FC = observer(() => {
           console.log('Connection closed');
         },
         onerror(err) {
-          commonStore.conversation[answerId].type = MessageType.Error;
-          commonStore.conversation[answerId].done = true;
+          commonStore.conversation[answerId!].type = MessageType.Error;
+          commonStore.conversation[answerId!].done = true;
           err = err.message || err;
           if (err && !err.includes('ReadableStreamDefaultReader'))
-            commonStore.conversation[answerId].content += '\n[ERROR]\n```\n' + err + '\n```';
+            commonStore.conversation[answerId!].content += '\n[ERROR]\n```\n' + err + '\n```';
           commonStore.setConversation(commonStore.conversation);
           commonStore.setConversationOrder([...commonStore.conversationOrder]);
           setTimeout(scrollToBottom);
           throw err;
         }
       });
-  };
+  }, []);
 
   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, index) => {
-          const messageItem = commonStore.conversation[uuid];
-          return <div
-            key={uuid}
-            className={classnames(
-              'flex gap-2 mb-2 overflow-hidden',
-              messageItem.side === 'left' ? 'flex-row' : 'flex-row-reverse'
-            )}
-            onMouseEnter={() => {
-              const utils = document.getElementById('utils-' + uuid);
-              if (utils) utils.classList.remove('invisible');
-            }}
-            onMouseLeave={() => {
-              const utils = document.getElementById('utils-' + uuid);
-              if (utils) utils.classList.add('invisible');
-            }}
-          >
-            <Avatar
-              color={messageItem.color}
-              name={messageItem.sender}
-              image={messageItem.avatarImg ? { src: messageItem.avatarImg } : undefined}
-            />
-            <div
-              className={classnames(
-                'p-2 rounded-lg overflow-hidden',
-                messageItem.side === 'left' ? 'bg-gray-200' : 'bg-blue-500',
-                messageItem.side === 'left' ? 'text-gray-600' : 'text-white'
-              )}
-            >
-              <MarkdownRender>{messageItem.content}</MarkdownRender>
-            </div>
-            <div className="flex flex-col gap-1 items-start">
-              <div className="grow" />
-              {(messageItem.type === MessageType.Error || !messageItem.done) &&
-                <PresenceBadge size="extra-small" status={
-                  messageItem.type === MessageType.Error ? 'busy' : 'away'
-                } />
-              }
-              <div className="flex invisible" id={'utils-' + uuid}>
-                <ReadButton content={messageItem.content} />
-                <CopyButton content={messageItem.content} />
-              </div>
-            </div>
-          </div>;
-        })}
+        {commonStore.conversationOrder.map(uuid =>
+          <ChatMessageItem key={uuid} uuid={uuid} onSubmit={onSubmit} />
+        )}
       </div>
       <div className="flex items-end gap-2">
         <DialogButton tooltip={t('Clear')}

frontend/src/pages/Completion.tsx

@@ -213,7 +213,6 @@ const CompletionPanel: FC = observer(() => {
         }),
         signal: completionSseController?.signal,
         onmessage(e) {
-          console.log('sse message', e);
           scrollToBottom();
           if (e.data === '[DONE]') {
             commonStore.setCompletionGenerating(false);
@@ -312,7 +311,7 @@ const CompletionPanel: FC = observer(() => {
           <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={-2} max={2}
+              <ValuedSlider value={params.presencePenalty} min={0} max={2}
                 step={0.1} input
                 onChange={(e, data) => {
                   setParams({
@@ -323,7 +322,7 @@ const CompletionPanel: FC = observer(() => {
           <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={-2} max={2}
+              <ValuedSlider value={params.frequencyPenalty} min={0} max={2}
                 step={0.1} input
                 onChange={(e, data) => {
                   setParams({

frontend/src/utils/index.tsx

@@ -126,19 +126,28 @@ export const getStrategy = (modelConfig: ModelConfig | undefined = undefined) =>
   let params: ModelParameters;
   if (modelConfig) params = modelConfig.modelParameters;
   else params = commonStore.getCurrentModelConfig().modelParameters;
+  const modelName = params.modelName.toLowerCase();
+  const avoidOverflow = params.precision !== 'fp32' && modelName.includes('world') && (modelName.includes('0.1b') || modelName.includes('0.4b') ||
+    modelName.includes('1.5b') || modelName.includes('1b5'));
   let strategy = '';
   switch (params.device) {
     case 'CPU':
+      if (avoidOverflow)
+        strategy = 'cpu fp32 *1 -> ';
       strategy += 'cpu ';
       strategy += params.precision === 'int8' ? 'fp32i8' : 'fp32';
       break;
     case 'CUDA':
+      if (avoidOverflow)
+        strategy = 'cuda fp32 *1 -> ';
       strategy += 'cuda ';
       strategy += params.precision === 'fp16' ? 'fp16' : params.precision === 'int8' ? 'fp16i8' : 'fp32';
       if (params.storedLayers < params.maxStoredLayers)
         strategy += ` *${params.storedLayers}+`;
       break;
     case 'MPS':
+      if (avoidOverflow)
+        strategy = 'mps fp32 *1 -> ';
       strategy += 'mps ';
       strategy += params.precision === 'fp16' ? 'fp16' : params.precision === 'int8' ? 'fp16i8' : 'fp32';
       break;
@@ -318,9 +327,10 @@ export function toastWithButton(text: string, buttonText: string, onClickButton:
 }
 
 export function getSupportedCustomCudaFile() {
-  if ([' 10', ' 16', ' 20', ' 30', 'MX', 'Tesla P', 'Quadro P', 'NVIDIA P', 'TITAN X', 'TITAN RTX', 'RTX A'].some(v => commonStore.status.device_name.includes(v)))
+  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)))
     return './backend-python/wkv_cuda_utils/wkv_cuda10_30.pyd';
-  else if ([' 40', 'RTX TITAN Ada'].some(v => commonStore.status.device_name.includes(v)))
+  else if ([' 40', 'RTX 5000 Ada', 'RTX 6000 Ada', 'RTX TITAN Ada', 'NVIDIA L40'].some(v => commonStore.status.device_name.includes(v)))
     return './backend-python/wkv_cuda_utils/wkv_cuda40.pyd';
   else
     return '';

manifest.json

@@ -1,5 +1,5 @@
 {
-  "version": "1.2.5",
+  "version": "1.2.7",
   "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）。"
@@ -18,8 +18,8 @@
     {
       "name": "RWKV-4-World-CHNtuned-0.1B-v1-20230617-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 0.1B v1 Enhanced Chinese",
-        "zh": "100+ 语言 0.1B v1 中文增强"
+        "en": "Global Languages 0.1B v1 Enhanced Chinese",
+        "zh": "全球语言 0.1B v1 中文增强"
       },
       "size": 385594610,
       "SHA256": "a3888f9958d378ee6d4976ae1c02edb698f4382e426086febafb4a69417b9080",
@@ -30,8 +30,8 @@
     {
       "name": "RWKV-4-World-0.1B-v1-20230520-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 0.1B v1",
-        "zh": "100+ 语言 0.1B v1"
+        "en": "Global Languages 0.1B v1",
+        "zh": "全球语言 0.1B v1"
       },
       "size": 385594610,
       "SHA256": "a10ef99df2a8f8a6801edf4fc92a9c49bedd63dcb900d3e5667a2136b3d671e7",
@@ -42,8 +42,8 @@
     {
       "name": "RWKV-4-World-CHNtuned-0.4B-v1-20230618-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 0.4B v1 Enhanced Chinese",
-        "zh": "100+ 语言 0.4B v1 中文增强"
+        "en": "Global Languages 0.4B v1 Enhanced Chinese",
+        "zh": "全球语言 0.4B v1 中文增强"
       },
       "size": 923362866,
       "SHA256": "dbd5302cbee596bbc900f97eb10b2af3001a7f2c7e4d8643bf8683b2cdbdd324",
@@ -54,8 +54,8 @@
     {
       "name": "RWKV-4-World-0.4B-v1-20230529-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 0.4B v1",
-        "zh": "100+ 语言 0.4B v1"
+        "en": "Global Languages 0.4B v1",
+        "zh": "全球语言 0.4B v1"
       },
       "size": 923362866,
       "SHA256": "4b4a2733cf5e5dc97dd62106f391d99895d16b11c5ccd10c89f28c52067a4919",
@@ -66,8 +66,8 @@
     {
       "name": "RWKV-4-World-CHNtuned-1.5B-v1-20230620-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 1.5B v1 Enhanced Chinese",
-        "zh": "100+ 语言 1.5B v1 中文增强"
+        "en": "Global Languages 1.5B v1 Enhanced Chinese",
+        "zh": "全球语言 1.5B v1 中文增强"
       },
       "size": 3155281586,
       "SHA256": "9f31f2ed5fe52dcf2d50208eb2efd764b9674dba2adb1baeff61997b4390a26b",
@@ -78,8 +78,8 @@
     {
       "name": "RWKV-4-World-1.5B-v1-OnlyForTest_57%_trained-20230529-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 1.5B v1 Test",
-        "zh": "100+ 语言 1.5B v1 测试"
+        "en": "Global Languages 1.5B v1 Test",
+        "zh": "全球语言 1.5B v1 测试"
       },
       "size": 3155281581,
       "SHA256": "ac36770931776c5aa179690918c9a3b0b5f4ebe3301ea3574a7e182209778788",
@@ -91,8 +91,8 @@
     {
       "name": "RWKV-4-World-1.5B-v1-OnlyForTest_81%_trained-20230603-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 1.5B v1 Test",
-        "zh": "100+ 语言 1.5B v1 测试"
+        "en": "Global Languages 1.5B v1 Test",
+        "zh": "全球语言 1.5B v1 测试"
       },
       "size": 3155281581,
       "SHA256": "044fb10daa71f4c012493ac8ef455c8c3301095b5f009dae58f0f6382a53e23c",
@@ -104,8 +104,8 @@
     {
       "name": "RWKV-4-World-1.5B-v1-20230607-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 1.5B v1",
-        "zh": "100+ 语言 1.5B v1"
+        "en": "Global Languages 1.5B v1",
+        "zh": "全球语言 1.5B v1"
       },
       "size": 3155281586,
       "SHA256": "05bad4ab0ce41250064153d5352587b83215a82eb50134489675129bd4ad1087",
@@ -117,8 +117,8 @@
     {
       "name": "RWKV-4-World-1.5B-v1-fixed-20230612-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 1.5B v1 fixed",
-        "zh": "100+ 语言 1.5B v1 修复"
+        "en": "Global Languages 1.5B v1 fixed",
+        "zh": "全球语言 1.5B v1 修复"
       },
       "size": 3155281586,
       "SHA256": "71f0c3229f9227cbcb8ae5fee6461197129a57e26366c4d23a49058417b046c9",
@@ -129,8 +129,8 @@
     {
       "name": "RWKV-4-World-3B-v1-OnlyForTest_35%_trained-20230529-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 3B v1 Test",
-        "zh": "100+ 语言 3B v1 测试"
+        "en": "Global Languages 3B v1 Test",
+        "zh": "全球语言 3B v1 测试"
       },
       "size": 6125597613,
       "SHA256": "e4ee6e91a80d56de43bc79841f3a8be3b7b215d7d9788f79c467b9b1f7f03cb8",
@@ -142,8 +142,8 @@
     {
       "name": "RWKV-4-World-3B-v1-OnlyForTest_52%_trained-20230603-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 3B v1 Test",
-        "zh": "100+ 语言 3B v1 测试"
+        "en": "Global Languages 3B v1 Test",
+        "zh": "全球语言 3B v1 测试"
       },
       "size": 6125597613,
       "SHA256": "aad3671078a0c686368add4f4b695a76c2ba1ddd505a64c0949bb003beeee9a3",
@@ -155,8 +155,8 @@
     {
       "name": "RWKV-4-World-3B-v1-OnlyForTest_64%_trained-20230607-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 3B v1 Test",
-        "zh": "100+ 语言 3B v1 测试"
+        "en": "Global Languages 3B v1 Test",
+        "zh": "全球语言 3B v1 测试"
       },
       "size": 6125597613,
       "SHA256": "49e8675e09e0786ca12a554442c37b9e809ed93e9211af937cd149968a6b81e9",
@@ -168,8 +168,8 @@
     {
       "name": "RWKV-4-World-3B-v1-OnlyForTest_80%_trained-20230612-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 3B v1 Test",
-        "zh": "100+ 语言 3B v1 测试"
+        "en": "Global Languages 3B v1 Test",
+        "zh": "全球语言 3B v1 测试"
       },
       "size": 6125597613,
       "SHA256": "3bb10caf3017871435d83f39facc8a729fd774020390153470f004eb3ef645bd",
@@ -181,8 +181,8 @@
     {
       "name": "RWKV-4-World-3B-v1-20230619-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 3B v1",
-        "zh": "100+ 语言 3B v1"
+        "en": "Global Languages 3B v1",
+        "zh": "全球语言 3B v1"
       },
       "size": 6125597618,
       "SHA256": "1b227af317fa25b6939ab3c7cd321226ca48b8fe4bbbd2df3db669f1482c54ba",
@@ -193,8 +193,8 @@
     {
       "name": "RWKV-4-World-7B-v1-OnlyForTest_30%_trained-20230529-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 7B v1 Test",
-        "zh": "100+ 语言 7B v1 测试"
+        "en": "Global Languages 7B v1 Test",
+        "zh": "全球语言 7B v1 测试"
       },
       "size": 15035393581,
       "SHA256": "05f91562b2ae8b025226e40b3fb536d6f8eb3c142ac899c0808ee1c9dc189ec4",
@@ -206,8 +206,8 @@
     {
       "name": "RWKV-4-World-7B-v1-OnlyForTest_40%_trained-20230601-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 7B v1 Test",
-        "zh": "100+ 语言 7B v1 测试"
+        "en": "Global Languages 7B v1 Test",
+        "zh": "全球语言 7B v1 测试"
       },
       "size": 15035393581,
       "SHA256": "63c060c472e45b6c3af2baaaee448ffd95f9b46e3cc6e1ef70ce7ecb1d01bcfa",
@@ -219,8 +219,8 @@
     {
       "name": "RWKV-4-World-7B-v1-OnlyForTest_52%_trained-20230606-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 7B v1 Test",
-        "zh": "100+ 语言 7B v1 测试"
+        "en": "Global Languages 7B v1 Test",
+        "zh": "全球语言 7B v1 测试"
       },
       "size": 15035393581,
       "SHA256": "636405626eadbab230e1a7dc2855bb6244e09b5850547dda7103f650b4849de7",
@@ -232,8 +232,8 @@
     {
       "name": "RWKV-4-World-7B-v1-OnlyForTest_64%_trained-20230610-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 7B v1 Test",
-        "zh": "100+ 语言 7B v1 测试"
+        "en": "Global Languages 7B v1 Test",
+        "zh": "全球语言 7B v1 测试"
       },
       "size": 15035393581,
       "SHA256": "8039be276f555318a5b2e9ad82b9d70001c12bd2e3e668048615fc7b09d5d9a4",
@@ -245,8 +245,8 @@
     {
       "name": "RWKV-4-World-7B-v1-OnlyForTest_75%_trained-20230615-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 7B v1 Test",
-        "zh": "100+ 语言 7B v1 测试"
+        "en": "Global Languages 7B v1 Test",
+        "zh": "全球语言 7B v1 测试"
       },
       "size": 15035393581,
       "SHA256": "a5f4246a18698a350a49988de7a8a01cbd765f8d11ee6427cabb93bf659f2d0d",
@@ -258,8 +258,8 @@
     {
       "name": "RWKV-4-World-7B-v1-OnlyForTest_84%_trained-20230618-ctx4096.pth",
       "desc": {
-        "en": "100+ Languages 7B v1 Test",
-        "zh": "100+ 语言 7B v1 测试"
+        "en": "Global Languages 7B v1 Test",
+        "zh": "全球语言 7B v1 测试"
       },
       "size": 15035393581,
       "SHA256": "dfb56e8ba32907cb47df83c8d702e7f350d9ad50a59b71b031da4681637588b3",