fix: estimate history token count

src/renderer/src/pages/home/components/Messages.tsx

@@ -2,10 +2,10 @@ import { useAssistant } from '@renderer/hooks/useAssistant'
 import { useProviderByAssistant } from '@renderer/hooks/useProvider'
 import { fetchChatCompletion, fetchMessagesSummary } from '@renderer/services/api'
 import { EVENT_NAMES, EventEmitter } from '@renderer/services/event'
-import { filterAtMessages } from '@renderer/services/message'
+import { estimateHistoryTokenCount, filterAtMessages } from '@renderer/services/messages'
 import LocalStorage from '@renderer/services/storage'
 import { Assistant, Message, Model, Topic } from '@renderer/types'
-import { estimateHistoryTokenCount, getBriefInfo, runAsyncFunction, uuid } from '@renderer/utils'
+import { getBriefInfo, runAsyncFunction, uuid } from '@renderer/utils'
 import { t } from 'i18next'
 import localforage from 'localforage'
 import { debounce, last, reverse } from 'lodash'

src/renderer/src/pages/home/components/input/Inputbar.tsx

@@ -13,10 +13,11 @@ import { useAssistant } from '@renderer/hooks/useAssistant'
 import { useSettings } from '@renderer/hooks/useSettings'
 import { getDefaultTopic } from '@renderer/services/assistant'
 import { EVENT_NAMES, EventEmitter } from '@renderer/services/event'
+import { estimateInputTokenCount } from '@renderer/services/messages'
 import store, { useAppSelector } from '@renderer/store'
 import { setGenerating } from '@renderer/store/runtime'
 import { Assistant, Message, Topic } from '@renderer/types'
-import { estimateInputTokenCount, uuid } from '@renderer/utils'
+import { uuid } from '@renderer/utils'
 import { Button, Popconfirm, Tag, Tooltip } from 'antd'
 import TextArea, { TextAreaRef } from 'antd/es/input/TextArea'
 import dayjs from 'dayjs'
@@ -32,8 +33,10 @@ interface Props {
   setActiveTopic: (topic: Topic) => void
 }
 
+let _text = ''
+
 const Inputbar: FC<Props> = ({ assistant, setActiveTopic }) => {
-  const [text, setText] = useState('')
+  const [text, setText] = useState(_text)
   const { addTopic } = useAssistant(assistant.id)
   const { sendMessageShortcut, showInputEstimatedTokens } = useSettings()
   const [expended, setExpend] = useState(false)
@@ -42,6 +45,8 @@ const Inputbar: FC<Props> = ({ assistant, setActiveTopic }) => {
   const inputRef = useRef<TextAreaRef>(null)
   const { t } = useTranslation()
 
+  _text = text
+
   const sendMessage = useCallback(() => {
     if (generating) {
       return

src/renderer/src/services/api.ts

@@ -14,7 +14,7 @@ import {
   getTranslateModel
 } from './assistant'
 import { EVENT_NAMES, EventEmitter } from './event'
-import { filterAtMessages } from './message'
+import { filterAtMessages } from './messages'
 import ProviderSDK from './ProviderSDK'
 
 export async function fetchChatCompletion({

src/renderer/src/services/message.ts

@@ -1,5 +0,0 @@
-import { Message } from '@renderer/types'
-
-export const filterAtMessages = (messages: Message[]) => {
-  return messages.filter((message) => message.type !== '@')
-}

src/renderer/src/services/messages.ts

@@ -0,0 +1,34 @@
+import { Assistant, Message } from '@renderer/types'
+import { getAssistantSettings } from '@renderer/utils'
+import { GPTTokens } from 'gpt-tokens'
+import { takeRight } from 'lodash'
+
+export const filterAtMessages = (messages: Message[]) => {
+  return messages.filter((message) => message.type !== '@')
+}
+
+export function estimateInputTokenCount(text: string) {
+  const input = new GPTTokens({
+    model: 'gpt-4o',
+    messages: [{ role: 'user', content: text }]
+  })
+
+  return input.usedTokens - 7
+}
+
+export function estimateHistoryTokenCount(assistant: Assistant, msgs: Message[]) {
+  const { contextCount } = getAssistantSettings(assistant)
+
+  const all = new GPTTokens({
+    model: 'gpt-4o',
+    messages: [
+      { role: 'system', content: assistant.prompt },
+      ...filterAtMessages(takeRight(msgs, contextCount)).map((message) => ({
+        role: message.role,
+        content: message.content
+      }))
+    ]
+  })
+
+  return all.usedTokens - 7
+}

src/renderer/src/utils/index.ts

@@ -1,8 +1,6 @@
 import { DEFAULT_CONEXTCOUNT, DEFAULT_TEMPERATURE } from '@renderer/config/constant'
-import { Assistant, AssistantSettings, Message, Model } from '@renderer/types'
+import { Assistant, AssistantSettings, Model } from '@renderer/types'
 import imageCompression from 'browser-image-compression'
-import { GPTTokens } from 'gpt-tokens'
-import { takeRight } from 'lodash'
 import { v4 as uuidv4 } from 'uuid'
 
 export const runAsyncFunction = async (fn: () => void) => {
@@ -181,29 +179,6 @@ export const getAssistantSettings = (assistant: Assistant): AssistantSettings =>
   }
 }
 
-export function estimateInputTokenCount(text: string) {
-  const input = new GPTTokens({
-    model: 'gpt-4o',
-    messages: [{ role: 'user', content: text }]
-  })
-
-  return input.usedTokens - 7
-}
-
-export function estimateHistoryTokenCount(assistant: Assistant, msgs: Message[]) {
-  const { contextCount } = getAssistantSettings(assistant)
-
-  const all = new GPTTokens({
-    model: 'gpt-4o',
-    messages: [
-      { role: 'system', content: assistant.prompt },
-      ...takeRight(msgs, contextCount).map((message) => ({ role: message.role, content: message.content }))
-    ]
-  })
-
-  return all.usedTokens - 7
-}
-
 /**
  * is valid proxy url
  * @param url proxy url