mirror of
https://github.com/CherryHQ/cherry-studio.git
synced 2025-12-24 18:50:56 +08:00
fix: message and rerank errors
commit 1c90e23d76a3e1008408bf29add122ccab7dbe6d Merge: 4e7920331fde0999Author: kangfenmao <kangfenmao@qq.com> Date: Sun May 11 18:43:56 2025 +0800 Merge branch 'main' into fix/next-release-bugs commit 4e7920336d838501b900a18f1c254f0d17fb54b8 Author: suyao <sy20010504@gmail.com> Date: Sun May 11 18:09:58 2025 +0800 refactor(GeminiProvider): implement image generation handling in chat responses commit cd1ce4c0c65bc2a111ddb3112722cb36e66b2515 Merge: 968de188235122c8Author: suyao <sy20010504@gmail.com> Date: Sun May 11 16:45:29 2025 +0800 Merge branch 'main' into fix/next-release-bugs commit 968de18866abacc55fd9bd74c8d618871e64ade6 Author: suyao <sy20010504@gmail.com> Date: Sun May 11 16:41:38 2025 +0800 fix: add new image generation models to the configuration commit 1eaf5801b4c0e2c3fa1aa2ed829b20d97ea57d3f Merge: cb76588de6655fffAuthor: suyao <sy20010504@gmail.com> Date: Sun May 11 13:14:17 2025 +0800 Merge branch 'main' into fix/next-release-bugs commit cb76588dc692f35da1f3d4fdbc9217c6a1a36501 Author: suyao <sy20010504@gmail.com> Date: Sun May 11 13:13:20 2025 +0800 fix: enhance error handling and metrics tracking across AI providers and message processing commit c2d6bdabc00b48419773d08c7d6630803d6310c8 Author: suyao <sy20010504@gmail.com> Date: Sun May 11 05:36:52 2025 +0800 fix: update regex for function calling models and improve time tracking logic commit 95340b87d0bba3cdcd173a181953afa42b26da9b Author: suyao <sy20010504@gmail.com> Date: Sun May 11 05:05:05 2025 +0800 fix: adjust thinking millisecond handling in message thunk commit f4d4d3901603f14df616582fa537f3d9c1a694eb Author: suyao <sy20010504@gmail.com> Date: Sun May 11 03:52:26 2025 +0800 fix: remove 'auto' option from qwen model supported options commit e26f603dfe1a9146b0575142363d5ceab30e32df Author: suyao <sy20010504@gmail.com> Date: Sun May 11 03:50:05 2025 +0800 fix: add support for inline base64 image data in image block commit bb0093c656b2b72158db1bf7bfef6aae46b8096c Merge: f9d1339bd39584fcAuthor: suyao <sy20010504@gmail.com> Date: Sun May 11 03:00:57 2025 +0800 Merge branch 'main' into fix/next-release-bugs commit f9d1339bd3367a4f80da23aac1fdc73b4cd2a075 Author: suyao <sy20010504@gmail.com> Date: Sun May 11 03:00:11 2025 +0800 fix: set default zoomFactor in settings commit 7cf6fd685662a012e2460e722edcbe5ed12f1a1c Merge: ba9c44823bebfe27Author: suyao <sy20010504@gmail.com> Date: Sun May 11 01:31:05 2025 +0800 Merge branch 'main' into fix/next-release-bugs commit ba9c44828578a3b5cc9fd1aaba80158615921785 Merge: 97dffe713bf0b6b3Author: kangfenmao <kangfenmao@qq.com> Date: Sat May 10 20:18:44 2025 +0800 Merge branch 'main' into fix/next-release-bugs commit 97dffe719ee00ace2325c80022a48c44f03d6e26 Author: lizhixuan <daoquqiexing@gmail.com> Date: Sat May 10 11:38:01 2025 +0800 feat(MessageMenubar): add edit option to dropdown for single message editing commit 70157439a1d9778cacc87b1781a84d159a8d3f33 Author: lizhixuan <daoquqiexing@gmail.com> Date: Sat May 10 10:53:10 2025 +0800 refactor(StreamProcessingService): comment out console.log for cleaner code commit fa33ba77a9306ad316f34da4149858192079f7a2 Author: lizhixuan <daoquqiexing@gmail.com> Date: Sat May 10 10:52:08 2025 +0800 refactor(messageThunk): remove console.log statements for cleaner code commit 6544c5d2990adf1943195e1d4d11383859a05488 Author: kangfenmao <kangfenmao@qq.com> Date: Sat May 10 10:17:44 2025 +0800 feat(i18n): add download success and failure messages in multiple languages commit e23bb6744a4a99b2062012691340f78fad4e1952 Merge: 55c5c55360cc1deeAuthor: kangfenmao <kangfenmao@qq.com> Date: Sat May 10 09:54:38 2025 +0800 Merge branch 'main' into fix/next-release-bugs commit 55c5c5533eab46a1de9c5511433ed1b1b9e90512 Author: suyao <sy20010504@gmail.com> Date: Fri May 9 22:19:35 2025 +0800 fix: update styled component props to use dollar sign prefix for consistency commit 7a5839e0efdf3eb648b3d26cdf7ce131a2821f6a Author: suyao <sy20010504@gmail.com> Date: Fri May 9 22:02:06 2025 +0800 fix: prevent default action in handleLinkClick for better link handling commit ecb075fddfc2bc5796a804ccde29aaa762d85da6 Merge: df149608963f04f7Author: suyao <sy20010504@gmail.com> Date: Fri May 9 21:55:54 2025 +0800 Merge branch 'main' into fix/next-release-bugs commit df149608904039903d74d7a72a722b6c1a567ee3 Merge: 93bd4eb9e29a9303Author: suyao <sy20010504@gmail.com> Date: Fri May 9 21:48:13 2025 +0800 Merge branch 'main' into fix/next-release-bugs commit 93bd4eb907816414af5c27efbaa2d3d014c707fc Merge: c4d1deb6 38ff9b90 Author: suyao <sy20010504@gmail.com> Date: Fri May 9 21:27:08 2025 +0800 Merge branch 'fix/next-release-bugs' of github.com:CherryHQ/cherry-studio into fix/next-release-bugs commit c4d1deb6911977a23cf731db6bed80b8352557ff Author: suyao <sy20010504@gmail.com> Date: Fri May 9 21:24:48 2025 +0800 feat: enhance citation handling and add metadata support in citation blocks commit 38ff9b90b8fee91ed1fba7b83c9470bc40bd3429 Author: MyPrototypeWhat <daoquqiexing@gmail.com> Date: Fri May 9 19:47:24 2025 +0800 fix: enhance logging and update async handling in StreamProcessingService and messageThunk - Enabled logging in `createStreamProcessor` for better debugging. - Added logging for updated messages in `updateExistingMessageAndBlocksInDB` and `saveUpdatesToDB`. - Updated `onTextComplete` and `onLLMWebSearchComplete` to handle asynchronous operations correctly. - Commented out unused `saveUpdatedBlockToDB` calls to prevent unnecessary database updates. commit cda0215c9c4e007c2c7240c3c9c8521fb7111774 Author: MyPrototypeWhat <daoquqiexing@gmail.com> Date: Fri May 9 18:47:55 2025 +0800 refactor: optimize block update logic and remove unused code - Updated `throttledBlockUpdate` to handle asynchronous updates directly. - Removed the unused `throttledBlockDbUpdate` function and its related logic. - Added cancellation for throttled updates on error and completion to improve performance and reliability. - Cleaned up commented-out code for better readability. commit de2f5b09c8384eabd4df7253047b838a2759671a Author: MyPrototypeWhat <daoquqiexing@gmail.com> Date: Fri May 9 18:42:00 2025 +0800 refactor: update message handling and state management - Simplified message editing logic by removing unnecessary success/error logging. - Added `updatedAt` timestamp to message updates for better tracking. - Refactored `editMessageBlocks` to accept message ID and updates directly. - Removed unused `getTopicLimit` function from `TopicManager`. - Updated message rendering to use `updatedAt` when available. - Enhanced type definitions to include `updatedAt` in message structure. commit 700fa13971cafb04314817d2d8732c8fbf33c9d7 Author: suyao <sy20010504@gmail.com> Date: Fri May 9 16:19:55 2025 +0800 Remove Zhipu mode and text-only link handling commit 06bd1338cd671b255e477cec76b12663ea759f4c Author: kangfenmao <kangfenmao@qq.com> Date: Fri May 9 15:49:02 2025 +0800 fix: update citation rendering logic in MainTextBlock component - Added a check to determine if the citation URL is a valid link. - Updated citation tag formatting to conditionally include the link based on the URL validity. commit e96c9a569f7708816f57505975d30667929eeb19 Author: kangfenmao <kangfenmao@qq.com> Date: Thu May 8 18:31:14 2025 +0800 style: update ChatNavigation and CitationsList components for improved UI consistency - Added header style to remove borders in ChatNavigation. - Enhanced CitationsList with new Skeleton loading state and improved layout for citation cards. - Refactored CitationLink to a div for better styling control and adjusted padding in OpenButton for a more polished appearance.
This commit is contained in:
kangfenmao
parent
1fde0999f8
commit
a64c8ded73
@ -189,7 +189,7 @@ export const TEXT_TO_IMAGE_REGEX = /flux|diffusion|stabilityai|sd-|dall|cogview|
|
||||
|
||||
// Reasoning models
|
||||
export const REASONING_REGEX =
|
||||
/^(o\d+(?:-[\w-]+)?|.*\b(?:reasoner|thinking)\b.*|.*-[rR]\d+.*|.*\bqwq(?:-[\w-]+)?\b.*|.*\bhunyuan-t1(?:-[\w-]+)?\b.*|.*\bglm-zero-preview\b.*|.*\bgrok-3-mini(?:-[\w-]+)?\b.*)$/i
|
||||
/^(o\d+(?:-[\w-]+)?|.*\b(?:reasoning|reasoner|thinking)\b.*|.*-[rR]\d+.*|.*\bqwq(?:-[\w-]+)?\b.*|.*\bhunyuan-t1(?:-[\w-]+)?\b.*|.*\bglm-zero-preview\b.*|.*\bgrok-3-mini(?:-[\w-]+)?\b.*)$/i
|
||||
|
||||
// Embedding models
|
||||
export const EMBEDDING_REGEX =
|
||||
@ -206,7 +206,7 @@ export const FUNCTION_CALLING_MODELS = [
|
||||
'gpt-4o-mini',
|
||||
'gpt-4',
|
||||
'gpt-4.5',
|
||||
'o1(?:-[\\w-]+)?',
|
||||
'o(1|3|4)(?:-[\\w-]+)?',
|
||||
'claude',
|
||||
'qwen',
|
||||
'qwen3',
|
||||
@ -2153,11 +2153,11 @@ export const TEXT_TO_IMAGES_MODELS_SUPPORT_IMAGE_ENHANCEMENT = [
|
||||
|
||||
export const GENERATE_IMAGE_MODELS = [
|
||||
'gemini-2.0-flash-exp-image-generation',
|
||||
'gemini-2.0-flash-preview-image-generation',
|
||||
'gemini-2.0-flash-exp',
|
||||
'grok-2-image-1212',
|
||||
'grok-2-image',
|
||||
'grok-2-image-latest',
|
||||
'gpt-4o-image',
|
||||
'gpt-image-1'
|
||||
]
|
||||
|
||||
@ -2172,6 +2172,7 @@ export const GEMINI_SEARCH_MODELS = [
|
||||
'gemini-2.5-pro-exp-03-25',
|
||||
'gemini-2.5-pro-preview',
|
||||
'gemini-2.5-pro-preview-03-25',
|
||||
'gemini-2.5-pro-preview-05-06',
|
||||
'gemini-2.5-flash-preview',
|
||||
'gemini-2.5-flash-preview-04-17'
|
||||
]
|
||||
|
||||
@ -35,7 +35,7 @@ const MODEL_SUPPORTED_OPTIONS: Record<string, ThinkingOption[]> = {
|
||||
default: ['off', 'low', 'medium', 'high'],
|
||||
grok: ['off', 'low', 'high'],
|
||||
gemini: ['off', 'low', 'medium', 'high', 'auto'],
|
||||
qwen: ['off', 'low', 'medium', 'high', 'auto']
|
||||
qwen: ['off', 'low', 'medium', 'high']
|
||||
}
|
||||
|
||||
// 选项转换映射表：当选项不支持时使用的替代选项
|
||||
|
||||
@ -16,7 +16,7 @@ const MessageErrorInfo: React.FC<{ block: ErrorMessageBlock }> = ({ block }) =>
|
||||
|
||||
const HTTP_ERROR_CODES = [400, 401, 403, 404, 429, 500, 502, 503, 504]
|
||||
if (block.error && HTTP_ERROR_CODES.includes(block.error?.status)) {
|
||||
return <Alert description={t(`error.http.${block.error.status}`)} type="error" />
|
||||
return <Alert description={t(`error.http.${block.error.status}`)} message={block.error?.message} type="error" />
|
||||
}
|
||||
if (block?.error?.message) {
|
||||
const errorKey = `error.${block.error.message}`
|
||||
|
||||
@ -4,7 +4,7 @@ import { MessageBlockStatus, type ThinkingMessageBlock } from '@renderer/types/n
|
||||
import { Collapse, message as antdMessage, Tooltip } from 'antd'
|
||||
import { Lightbulb } from 'lucide-react'
|
||||
import { motion } from 'motion/react'
|
||||
import { memo, useCallback, useEffect, useMemo, useState } from 'react'
|
||||
import { memo, useCallback, useEffect, useMemo, useRef, useState } from 'react'
|
||||
import { useTranslation } from 'react-i18next'
|
||||
import styled from 'styled-components'
|
||||
|
||||
@ -40,6 +40,8 @@ const ThinkingBlock: React.FC<Props> = ({ block }) => {
|
||||
const { t } = useTranslation()
|
||||
const { messageFont, fontSize, thoughtAutoCollapse } = useSettings()
|
||||
const [activeKey, setActiveKey] = useState<'thought' | ''>(thoughtAutoCollapse ? '' : 'thought')
|
||||
const [thinkingTime, setThinkingTime] = useState(block.thinking_millsec || 0)
|
||||
const intervalId = useRef<NodeJS.Timeout>(null)
|
||||
|
||||
const isThinking = useMemo(() => block.status === MessageBlockStatus.STREAMING, [block.status])
|
||||
|
||||
@ -73,13 +75,31 @@ const ThinkingBlock: React.FC<Props> = ({ block }) => {
|
||||
}
|
||||
}, [block.content, t])
|
||||
|
||||
// FIXME: 这里统计的和请求处统计的有一定误差
|
||||
useEffect(() => {
|
||||
if (isThinking) {
|
||||
intervalId.current = setInterval(() => {
|
||||
setThinkingTime((prev) => prev + 100)
|
||||
}, 100)
|
||||
} else if (intervalId.current) {
|
||||
// 立即清除计时器
|
||||
clearInterval(intervalId.current)
|
||||
intervalId.current = null
|
||||
}
|
||||
|
||||
return () => {
|
||||
if (intervalId.current) {
|
||||
window.clearInterval(intervalId.current)
|
||||
}
|
||||
}
|
||||
}, [isThinking])
|
||||
|
||||
const thinkingTimeSeconds = useMemo(() => (thinkingTime / 1000).toFixed(1), [thinkingTime])
|
||||
|
||||
if (!block.content) {
|
||||
return null
|
||||
}
|
||||
|
||||
const thinkingTime = block.thinking_millsec || 0
|
||||
const thinkingTimeSeconds = (thinkingTime / 1000).toFixed(1)
|
||||
|
||||
return (
|
||||
<CollapseContainer
|
||||
activeKey={activeKey}
|
||||
|
||||
@ -10,35 +10,6 @@ interface Props {
|
||||
}
|
||||
|
||||
const MessageContent: React.FC<Props> = ({ message }) => {
|
||||
// const { t } = useTranslation()
|
||||
// if (message.status === 'pending') {
|
||||
// return (
|
||||
|
||||
// )
|
||||
// }
|
||||
|
||||
// if (message.status === 'searching') {
|
||||
// return (
|
||||
// <SearchingContainer>
|
||||
// <Search size={24} />
|
||||
// <SearchingText>{t('message.searching')}</SearchingText>
|
||||
// <BarLoader color="#1677ff" />
|
||||
// </SearchingContainer>
|
||||
// )
|
||||
// }
|
||||
|
||||
// if (message.status === 'error') {
|
||||
// return <MessageError message={message} />
|
||||
// }
|
||||
|
||||
// if (message.type === '@' && model) {
|
||||
// const content = `[@${model.name}](#) ${getBriefInfo(message.content)}`
|
||||
// return <Markdown message={{ ...message, content }} />
|
||||
// }
|
||||
// const toolUseRegex = /<tool_use>([\s\S]*?)<\/tool_use>/g
|
||||
|
||||
// console.log('message', message)
|
||||
|
||||
return (
|
||||
<>
|
||||
<Flex gap="8px" wrap style={{ marginBottom: 10 }}>
|
||||
|
||||
@ -1,45 +0,0 @@
|
||||
import type { ErrorMessageBlock } from '@renderer/types/newMessage'
|
||||
import { Alert as AntdAlert } from 'antd'
|
||||
import { FC } from 'react'
|
||||
import { useTranslation } from 'react-i18next'
|
||||
import styled from 'styled-components'
|
||||
|
||||
const MessageError: FC<{ block: ErrorMessageBlock }> = ({ block }) => {
|
||||
return (
|
||||
<>
|
||||
{/* <Markdown block={block} role={role} />
|
||||
{block.error && (
|
||||
<Markdown
|
||||
message={{
|
||||
...block,
|
||||
content: formatErrorMessage(block.error)
|
||||
}}
|
||||
/>
|
||||
)} */}
|
||||
<MessageErrorInfo block={block} />
|
||||
</>
|
||||
)
|
||||
}
|
||||
|
||||
const MessageErrorInfo: FC<{ block: ErrorMessageBlock }> = ({ block }) => {
|
||||
const { t } = useTranslation()
|
||||
|
||||
const HTTP_ERROR_CODES = [400, 401, 403, 404, 429, 500, 502, 503, 504]
|
||||
console.log('block', block)
|
||||
if (block.error && HTTP_ERROR_CODES.includes(block.error?.status)) {
|
||||
return <Alert description={t(`error.http.${block.error.status}`)} type="error" />
|
||||
}
|
||||
if (block?.error?.message) {
|
||||
return <Alert description={block.error.message} type="error" />
|
||||
}
|
||||
|
||||
return <Alert description={t('error.chat.response')} type="error" />
|
||||
}
|
||||
|
||||
const Alert = styled(AntdAlert)`
|
||||
margin: 15px 0 8px;
|
||||
padding: 10px;
|
||||
font-size: 12px;
|
||||
`
|
||||
|
||||
export default MessageError
|
||||
@ -30,10 +30,12 @@ import {
|
||||
MCPCallToolResponse,
|
||||
MCPTool,
|
||||
MCPToolResponse,
|
||||
Metrics,
|
||||
Model,
|
||||
Provider,
|
||||
Suggestion,
|
||||
ToolCallResponse,
|
||||
Usage,
|
||||
WebSearchSource
|
||||
} from '@renderer/types'
|
||||
import { ChunkType } from '@renderer/types/chunk'
|
||||
@ -47,7 +49,7 @@ import {
|
||||
} from '@renderer/utils/mcp-tools'
|
||||
import { findFileBlocks, findImageBlocks, getMainTextContent } from '@renderer/utils/messageUtils/find'
|
||||
import { buildSystemPrompt } from '@renderer/utils/prompt'
|
||||
import { first, flatten, sum, takeRight } from 'lodash'
|
||||
import { first, flatten, takeRight } from 'lodash'
|
||||
import OpenAI from 'openai'
|
||||
|
||||
import { CompletionsParams } from '.'
|
||||
@ -270,77 +272,82 @@ export default class AnthropicProvider extends BaseProvider {
|
||||
...this.getCustomParameters(assistant)
|
||||
}
|
||||
|
||||
let time_first_token_millsec = 0
|
||||
let time_first_content_millsec = 0
|
||||
let checkThinkingContent = false
|
||||
let thinking_content = ''
|
||||
const start_time_millsec = new Date().getTime()
|
||||
|
||||
if (!streamOutput) {
|
||||
const message = await this.sdk.messages.create({ ...body, stream: false })
|
||||
const time_completion_millsec = new Date().getTime() - start_time_millsec
|
||||
|
||||
let text = ''
|
||||
let reasoning_content = ''
|
||||
|
||||
if (message.content && message.content.length > 0) {
|
||||
const thinkingBlock = message.content.find((block) => block.type === 'thinking')
|
||||
const textBlock = message.content.find((block) => block.type === 'text')
|
||||
|
||||
if (thinkingBlock && 'thinking' in thinkingBlock) {
|
||||
reasoning_content = thinkingBlock.thinking
|
||||
}
|
||||
|
||||
if (textBlock && 'text' in textBlock) {
|
||||
text = textBlock.text
|
||||
}
|
||||
}
|
||||
|
||||
return onChunk({
|
||||
type: ChunkType.BLOCK_COMPLETE,
|
||||
response: {
|
||||
text,
|
||||
reasoning_content,
|
||||
usage: message.usage as any,
|
||||
metrics: {
|
||||
completion_tokens: message.usage.output_tokens,
|
||||
time_completion_millsec,
|
||||
time_first_token_millsec: 0
|
||||
}
|
||||
}
|
||||
})
|
||||
}
|
||||
|
||||
const { abortController, cleanup } = this.createAbortController(lastUserMessage?.id)
|
||||
const { signal } = abortController
|
||||
|
||||
const finalUsage: Usage = {
|
||||
completion_tokens: 0,
|
||||
prompt_tokens: 0,
|
||||
total_tokens: 0
|
||||
}
|
||||
|
||||
const finalMetrics: Metrics = {
|
||||
completion_tokens: 0,
|
||||
time_completion_millsec: 0,
|
||||
time_first_token_millsec: 0
|
||||
}
|
||||
const toolResponses: MCPToolResponse[] = []
|
||||
|
||||
const processStream = (body: MessageCreateParamsNonStreaming, idx: number) => {
|
||||
const processStream = async (body: MessageCreateParamsNonStreaming, idx: number) => {
|
||||
let time_first_token_millsec = 0
|
||||
const start_time_millsec = new Date().getTime()
|
||||
|
||||
if (!streamOutput) {
|
||||
const message = await this.sdk.messages.create({ ...body, stream: false })
|
||||
const time_completion_millsec = new Date().getTime() - start_time_millsec
|
||||
|
||||
let text = ''
|
||||
let reasoning_content = ''
|
||||
|
||||
if (message.content && message.content.length > 0) {
|
||||
const thinkingBlock = message.content.find((block) => block.type === 'thinking')
|
||||
const textBlock = message.content.find((block) => block.type === 'text')
|
||||
|
||||
if (thinkingBlock && 'thinking' in thinkingBlock) {
|
||||
reasoning_content = thinkingBlock.thinking
|
||||
}
|
||||
|
||||
if (textBlock && 'text' in textBlock) {
|
||||
text = textBlock.text
|
||||
}
|
||||
}
|
||||
|
||||
return onChunk({
|
||||
type: ChunkType.BLOCK_COMPLETE,
|
||||
response: {
|
||||
text,
|
||||
reasoning_content,
|
||||
usage: message.usage as any,
|
||||
metrics: {
|
||||
completion_tokens: message.usage.output_tokens,
|
||||
time_completion_millsec,
|
||||
time_first_token_millsec: 0
|
||||
}
|
||||
}
|
||||
})
|
||||
}
|
||||
|
||||
let thinking_content = ''
|
||||
let isFirstChunk = true
|
||||
|
||||
return new Promise<void>((resolve, reject) => {
|
||||
// 等待接口返回流
|
||||
const toolCalls: ToolUseBlock[] = []
|
||||
let hasThinkingContent = false
|
||||
|
||||
this.sdk.messages
|
||||
.stream({ ...body, stream: true }, { signal, timeout: 5 * 60 * 1000 })
|
||||
.on('text', (text) => {
|
||||
if (hasThinkingContent && !checkThinkingContent) {
|
||||
checkThinkingContent = true
|
||||
onChunk({
|
||||
type: ChunkType.THINKING_COMPLETE,
|
||||
text: thinking_content,
|
||||
thinking_millsec: new Date().getTime() - time_first_content_millsec
|
||||
})
|
||||
}
|
||||
if (time_first_token_millsec == 0) {
|
||||
time_first_token_millsec = new Date().getTime()
|
||||
}
|
||||
|
||||
thinking_content = ''
|
||||
checkThinkingContent = false
|
||||
hasThinkingContent = false
|
||||
|
||||
if (!hasThinkingContent && time_first_content_millsec === 0) {
|
||||
time_first_content_millsec = new Date().getTime()
|
||||
if (isFirstChunk) {
|
||||
isFirstChunk = false
|
||||
if (time_first_token_millsec == 0) {
|
||||
time_first_token_millsec = new Date().getTime()
|
||||
} else {
|
||||
onChunk({
|
||||
type: ChunkType.THINKING_COMPLETE,
|
||||
text: thinking_content,
|
||||
thinking_millsec: new Date().getTime() - time_first_token_millsec
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
onChunk({ type: ChunkType.TEXT_DELTA, text })
|
||||
@ -372,34 +379,22 @@ export default class AnthropicProvider extends BaseProvider {
|
||||
})
|
||||
}
|
||||
}
|
||||
if (block.type === 'tool_use') {
|
||||
toolCalls.push(block)
|
||||
}
|
||||
})
|
||||
.on('thinking', (thinking) => {
|
||||
hasThinkingContent = true
|
||||
const currentTime = new Date().getTime() // Get current time for each chunk
|
||||
|
||||
if (time_first_token_millsec == 0) {
|
||||
time_first_token_millsec = currentTime
|
||||
time_first_token_millsec = new Date().getTime()
|
||||
}
|
||||
|
||||
// Set time_first_content_millsec ONLY when the first content (thinking or text) arrives
|
||||
if (time_first_content_millsec === 0) {
|
||||
time_first_content_millsec = currentTime
|
||||
}
|
||||
|
||||
// Calculate thinking time as time elapsed since start until this chunk
|
||||
const thinking_time = currentTime - time_first_content_millsec
|
||||
onChunk({
|
||||
type: ChunkType.THINKING_DELTA,
|
||||
text: thinking,
|
||||
thinking_millsec: thinking_time
|
||||
thinking_millsec: new Date().getTime() - time_first_token_millsec
|
||||
})
|
||||
thinking_content += thinking
|
||||
})
|
||||
.on('contentBlock', (content) => {
|
||||
if (content.type === 'tool_use') {
|
||||
toolCalls.push(content)
|
||||
}
|
||||
})
|
||||
.on('finalMessage', async (message) => {
|
||||
const toolResults: Awaited<ReturnType<typeof parseAndCallTools>> = []
|
||||
// tool call
|
||||
@ -458,29 +453,28 @@ export default class AnthropicProvider extends BaseProvider {
|
||||
newBody.messages = userMessages
|
||||
|
||||
onChunk({ type: ChunkType.LLM_RESPONSE_CREATED })
|
||||
await processStream(newBody, idx + 1)
|
||||
try {
|
||||
await processStream(newBody, idx + 1)
|
||||
} catch (error) {
|
||||
console.error('Error processing stream:', error)
|
||||
reject(error)
|
||||
}
|
||||
}
|
||||
|
||||
const time_completion_millsec = new Date().getTime() - start_time_millsec
|
||||
finalUsage.prompt_tokens += message.usage.input_tokens
|
||||
finalUsage.completion_tokens += message.usage.output_tokens
|
||||
finalUsage.total_tokens += finalUsage.prompt_tokens + finalUsage.completion_tokens
|
||||
finalMetrics.completion_tokens = finalUsage.completion_tokens
|
||||
finalMetrics.time_completion_millsec += new Date().getTime() - start_time_millsec
|
||||
finalMetrics.time_first_token_millsec = time_first_token_millsec - start_time_millsec
|
||||
|
||||
onChunk({
|
||||
type: ChunkType.BLOCK_COMPLETE,
|
||||
response: {
|
||||
usage: {
|
||||
prompt_tokens: message.usage.input_tokens,
|
||||
completion_tokens: message.usage.output_tokens,
|
||||
total_tokens: sum(Object.values(message.usage))
|
||||
},
|
||||
metrics: {
|
||||
completion_tokens: message.usage.output_tokens,
|
||||
time_completion_millsec,
|
||||
time_first_token_millsec: time_first_token_millsec - start_time_millsec
|
||||
}
|
||||
usage: finalUsage,
|
||||
metrics: finalMetrics
|
||||
}
|
||||
})
|
||||
// FIXME: 临时方案，重置时间戳和思考内容
|
||||
time_first_token_millsec = 0
|
||||
time_first_content_millsec = 0
|
||||
resolve()
|
||||
})
|
||||
.on('error', (error) => reject(error))
|
||||
|
||||
@ -40,6 +40,7 @@ import {
|
||||
MCPCallToolResponse,
|
||||
MCPTool,
|
||||
MCPToolResponse,
|
||||
Metrics,
|
||||
Model,
|
||||
Provider,
|
||||
Suggestion,
|
||||
@ -126,7 +127,6 @@ export default class GeminiProvider extends BaseProvider {
|
||||
* @returns The message contents
|
||||
*/
|
||||
private async getMessageContents(message: Message): Promise<Content> {
|
||||
console.log('getMessageContents', message)
|
||||
const role = message.role === 'user' ? 'user' : 'model'
|
||||
const parts: Part[] = [{ text: await this.getMessageContent(message) }]
|
||||
// Add any generated images from previous responses
|
||||
@ -153,6 +153,16 @@ export default class GeminiProvider extends BaseProvider {
|
||||
}
|
||||
}
|
||||
}
|
||||
const file = imageBlock.file
|
||||
if (file) {
|
||||
const base64Data = await window.api.file.base64Image(file.id + file.ext)
|
||||
parts.push({
|
||||
inlineData: {
|
||||
data: base64Data.base64,
|
||||
mimeType: base64Data.mime
|
||||
} as Part['inlineData']
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
const fileBlocks = findFileBlocks(message)
|
||||
@ -186,6 +196,50 @@ export default class GeminiProvider extends BaseProvider {
|
||||
}
|
||||
}
|
||||
|
||||
private async getImageFileContents(message: Message): Promise<Content> {
|
||||
const role = message.role === 'user' ? 'user' : 'model'
|
||||
const content = getMainTextContent(message)
|
||||
const parts: Part[] = [{ text: content }]
|
||||
const imageBlocks = findImageBlocks(message)
|
||||
for (const imageBlock of imageBlocks) {
|
||||
if (
|
||||
imageBlock.metadata?.generateImageResponse?.images &&
|
||||
imageBlock.metadata.generateImageResponse.images.length > 0
|
||||
) {
|
||||
for (const imageUrl of imageBlock.metadata.generateImageResponse.images) {
|
||||
if (imageUrl && imageUrl.startsWith('data:')) {
|
||||
// Extract base64 data and mime type from the data URL
|
||||
const matches = imageUrl.match(/^data:(.+);base64,(.*)$/)
|
||||
if (matches && matches.length === 3) {
|
||||
const mimeType = matches[1]
|
||||
const base64Data = matches[2]
|
||||
parts.push({
|
||||
inlineData: {
|
||||
data: base64Data,
|
||||
mimeType: mimeType
|
||||
} as Part['inlineData']
|
||||
})
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
const file = imageBlock.file
|
||||
if (file) {
|
||||
const base64Data = await window.api.file.base64Image(file.id + file.ext)
|
||||
parts.push({
|
||||
inlineData: {
|
||||
data: base64Data.base64,
|
||||
mimeType: base64Data.mime
|
||||
} as Part['inlineData']
|
||||
})
|
||||
}
|
||||
}
|
||||
return {
|
||||
role,
|
||||
parts: parts
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Get the safety settings
|
||||
* @returns The safety settings
|
||||
@ -273,6 +327,18 @@ export default class GeminiProvider extends BaseProvider {
|
||||
}: CompletionsParams): Promise<void> {
|
||||
const defaultModel = getDefaultModel()
|
||||
const model = assistant.model || defaultModel
|
||||
let canGenerateImage = false
|
||||
if (isGenerateImageModel(model)) {
|
||||
if (model.id === 'gemini-2.0-flash-exp') {
|
||||
canGenerateImage = assistant.enableGenerateImage!
|
||||
} else {
|
||||
canGenerateImage = true
|
||||
}
|
||||
}
|
||||
if (canGenerateImage) {
|
||||
await this.generateImageByChat({ messages, assistant, onChunk })
|
||||
return
|
||||
}
|
||||
const { contextCount, maxTokens, streamOutput, enableToolUse } = getAssistantSettings(assistant)
|
||||
|
||||
const userMessages = filterUserRoleStartMessages(
|
||||
@ -309,21 +375,10 @@ export default class GeminiProvider extends BaseProvider {
|
||||
})
|
||||
}
|
||||
|
||||
let canGenerateImage = false
|
||||
if (isGenerateImageModel(model)) {
|
||||
if (model.id === 'gemini-2.0-flash-exp') {
|
||||
canGenerateImage = assistant.enableGenerateImage!
|
||||
} else {
|
||||
canGenerateImage = true
|
||||
}
|
||||
}
|
||||
|
||||
const generateContentConfig: GenerateContentConfig = {
|
||||
responseModalities: canGenerateImage ? [Modality.TEXT, Modality.IMAGE] : undefined,
|
||||
responseMimeType: canGenerateImage ? 'text/plain' : undefined,
|
||||
safetySettings: this.getSafetySettings(),
|
||||
// generate image don't need system instruction
|
||||
systemInstruction: isGemmaModel(model) || canGenerateImage ? undefined : systemInstruction,
|
||||
systemInstruction: isGemmaModel(model) ? undefined : systemInstruction,
|
||||
temperature: assistant?.settings?.temperature,
|
||||
topP: assistant?.settings?.topP,
|
||||
maxOutputTokens: maxTokens,
|
||||
@ -360,8 +415,17 @@ export default class GeminiProvider extends BaseProvider {
|
||||
}
|
||||
}
|
||||
|
||||
const start_time_millsec = new Date().getTime()
|
||||
let time_first_token_millsec = 0
|
||||
const finalUsage: Usage = {
|
||||
completion_tokens: 0,
|
||||
prompt_tokens: 0,
|
||||
total_tokens: 0
|
||||
}
|
||||
|
||||
const finalMetrics: Metrics = {
|
||||
completion_tokens: 0,
|
||||
time_completion_millsec: 0,
|
||||
time_first_token_millsec: 0
|
||||
}
|
||||
|
||||
const { cleanup, abortController } = this.createAbortController(userLastMessage?.id, true)
|
||||
|
||||
@ -435,6 +499,8 @@ export default class GeminiProvider extends BaseProvider {
|
||||
history.push(messageContents)
|
||||
|
||||
let functionCalls: FunctionCall[] = []
|
||||
let time_first_token_millsec = 0
|
||||
const start_time_millsec = new Date().getTime()
|
||||
|
||||
if (stream instanceof GenerateContentResponse) {
|
||||
let content = ''
|
||||
@ -494,45 +560,28 @@ export default class GeminiProvider extends BaseProvider {
|
||||
} as BlockCompleteChunk)
|
||||
} else {
|
||||
let content = ''
|
||||
let final_time_completion_millsec = 0
|
||||
let lastUsage: Usage | undefined = undefined
|
||||
for await (const chunk of stream) {
|
||||
if (window.keyv.get(EVENT_NAMES.CHAT_COMPLETION_PAUSED)) break
|
||||
|
||||
// --- Calculate Metrics ---
|
||||
if (time_first_token_millsec == 0 && chunk.text !== undefined) {
|
||||
// Update based on text arrival
|
||||
time_first_token_millsec = new Date().getTime() - start_time_millsec
|
||||
if (time_first_token_millsec == 0) {
|
||||
time_first_token_millsec = new Date().getTime()
|
||||
}
|
||||
|
||||
// 1. Text Content
|
||||
if (chunk.text !== undefined) {
|
||||
content += chunk.text
|
||||
onChunk({ type: ChunkType.TEXT_DELTA, text: chunk.text })
|
||||
}
|
||||
|
||||
// 2. Usage Data
|
||||
if (chunk.usageMetadata) {
|
||||
lastUsage = {
|
||||
prompt_tokens: chunk.usageMetadata.promptTokenCount || 0,
|
||||
completion_tokens: chunk.usageMetadata.candidatesTokenCount || 0,
|
||||
total_tokens: chunk.usageMetadata.totalTokenCount || 0
|
||||
}
|
||||
final_time_completion_millsec = new Date().getTime() - start_time_millsec
|
||||
}
|
||||
|
||||
// 4. Image Generation
|
||||
const generateImage = this.processGeminiImageResponse(chunk, onChunk)
|
||||
if (generateImage?.images?.length) {
|
||||
onChunk({ type: ChunkType.IMAGE_COMPLETE, image: generateImage })
|
||||
}
|
||||
|
||||
if (chunk.candidates?.[0]?.finishReason) {
|
||||
if (chunk.text) {
|
||||
onChunk({ type: ChunkType.TEXT_COMPLETE, text: content })
|
||||
}
|
||||
if (chunk.usageMetadata) {
|
||||
finalUsage.prompt_tokens += chunk.usageMetadata.promptTokenCount || 0
|
||||
finalUsage.completion_tokens += chunk.usageMetadata.candidatesTokenCount || 0
|
||||
finalUsage.total_tokens += chunk.usageMetadata.totalTokenCount || 0
|
||||
}
|
||||
if (chunk.candidates?.[0]?.groundingMetadata) {
|
||||
// 3. Grounding/Search Metadata
|
||||
const groundingMetadata = chunk.candidates?.[0]?.groundingMetadata
|
||||
onChunk({
|
||||
type: ChunkType.LLM_WEB_SEARCH_COMPLETE,
|
||||
@ -551,35 +600,37 @@ export default class GeminiProvider extends BaseProvider {
|
||||
functionCalls = functionCalls.concat(chunk.functionCalls)
|
||||
}
|
||||
|
||||
onChunk({
|
||||
type: ChunkType.BLOCK_COMPLETE,
|
||||
response: {
|
||||
metrics: {
|
||||
completion_tokens: lastUsage?.completion_tokens,
|
||||
time_completion_millsec: final_time_completion_millsec,
|
||||
time_first_token_millsec
|
||||
},
|
||||
usage: lastUsage
|
||||
}
|
||||
})
|
||||
}
|
||||
|
||||
// --- End Incremental onChunk calls ---
|
||||
|
||||
// Call processToolUses AFTER potentially processing text content in this chunk
|
||||
// This assumes tools might be specified within the text stream
|
||||
// Note: parseAndCallTools inside should handle its own onChunk for tool responses
|
||||
let toolResults: Awaited<ReturnType<typeof parseAndCallTools>> = []
|
||||
if (functionCalls.length) {
|
||||
toolResults = await processToolCalls(functionCalls)
|
||||
}
|
||||
if (content.length) {
|
||||
toolResults = toolResults.concat(await processToolUses(content))
|
||||
}
|
||||
if (toolResults.length) {
|
||||
await processToolResults(toolResults, idx)
|
||||
finalMetrics.completion_tokens = finalUsage.completion_tokens
|
||||
finalMetrics.time_completion_millsec += new Date().getTime() - start_time_millsec
|
||||
finalMetrics.time_first_token_millsec =
|
||||
(finalMetrics.time_first_token_millsec || 0) + (time_first_token_millsec - start_time_millsec)
|
||||
}
|
||||
}
|
||||
|
||||
// --- End Incremental onChunk calls ---
|
||||
|
||||
// Call processToolUses AFTER potentially processing text content in this chunk
|
||||
// This assumes tools might be specified within the text stream
|
||||
// Note: parseAndCallTools inside should handle its own onChunk for tool responses
|
||||
let toolResults: Awaited<ReturnType<typeof parseAndCallTools>> = []
|
||||
if (functionCalls.length) {
|
||||
toolResults = await processToolCalls(functionCalls)
|
||||
}
|
||||
if (content.length) {
|
||||
toolResults = toolResults.concat(await processToolUses(content))
|
||||
}
|
||||
if (toolResults.length) {
|
||||
await processToolResults(toolResults, idx)
|
||||
}
|
||||
|
||||
// FIXME: 由于递归，会发送n次
|
||||
onChunk({
|
||||
type: ChunkType.BLOCK_COMPLETE,
|
||||
response: {
|
||||
usage: finalUsage,
|
||||
metrics: finalMetrics
|
||||
}
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
@ -605,17 +656,6 @@ export default class GeminiProvider extends BaseProvider {
|
||||
})
|
||||
|
||||
await processStream(userMessagesStream, 0).finally(cleanup)
|
||||
|
||||
const final_time_completion_millsec = new Date().getTime() - start_time_millsec
|
||||
onChunk({
|
||||
type: ChunkType.BLOCK_COMPLETE,
|
||||
response: {
|
||||
metrics: {
|
||||
time_completion_millsec: final_time_completion_millsec,
|
||||
time_first_token_millsec
|
||||
}
|
||||
}
|
||||
})
|
||||
}
|
||||
|
||||
/**
|
||||
@ -949,8 +989,97 @@ export default class GeminiProvider extends BaseProvider {
|
||||
return data.embeddings?.[0]?.values?.length || 0
|
||||
}
|
||||
|
||||
public generateImageByChat(): Promise<void> {
|
||||
throw new Error('Method not implemented.')
|
||||
public async generateImageByChat({ messages, assistant, onChunk }): Promise<void> {
|
||||
const defaultModel = getDefaultModel()
|
||||
const model = assistant.model || defaultModel
|
||||
const { contextCount, maxTokens } = getAssistantSettings(assistant)
|
||||
const userMessages = filterUserRoleStartMessages(
|
||||
filterEmptyMessages(filterContextMessages(takeRight(messages, contextCount + 2)))
|
||||
)
|
||||
|
||||
const userLastMessage = userMessages.pop()
|
||||
const { abortController } = this.createAbortController(userLastMessage?.id, true)
|
||||
const { signal } = abortController
|
||||
const generateContentConfig: GenerateContentConfig = {
|
||||
responseModalities: [Modality.TEXT, Modality.IMAGE],
|
||||
responseMimeType: 'text/plain',
|
||||
safetySettings: this.getSafetySettings(),
|
||||
temperature: assistant?.settings?.temperature,
|
||||
topP: assistant?.settings?.top_p,
|
||||
maxOutputTokens: maxTokens,
|
||||
abortSignal: signal,
|
||||
...this.getCustomParameters(assistant)
|
||||
}
|
||||
const history: Content[] = []
|
||||
try {
|
||||
for (const message of userMessages) {
|
||||
history.push(await this.getImageFileContents(message))
|
||||
}
|
||||
|
||||
let time_first_token_millsec = 0
|
||||
const start_time_millsec = new Date().getTime()
|
||||
onChunk({ type: ChunkType.LLM_RESPONSE_CREATED })
|
||||
const chat = this.sdk.chats.create({
|
||||
model: model.id,
|
||||
config: generateContentConfig,
|
||||
history: history
|
||||
})
|
||||
let content = ''
|
||||
const finalUsage: Usage = {
|
||||
prompt_tokens: 0,
|
||||
completion_tokens: 0,
|
||||
total_tokens: 0
|
||||
}
|
||||
const userMessage: Content = await this.getImageFileContents(userLastMessage!)
|
||||
const response = await chat.sendMessageStream({
|
||||
message: userMessage.parts!,
|
||||
config: {
|
||||
...generateContentConfig,
|
||||
abortSignal: signal
|
||||
}
|
||||
})
|
||||
for await (const chunk of response as AsyncGenerator<GenerateContentResponse>) {
|
||||
if (time_first_token_millsec == 0) {
|
||||
time_first_token_millsec = new Date().getTime()
|
||||
}
|
||||
|
||||
if (chunk.text !== undefined) {
|
||||
content += chunk.text
|
||||
onChunk({ type: ChunkType.TEXT_DELTA, text: chunk.text })
|
||||
}
|
||||
const generateImage = this.processGeminiImageResponse(chunk, onChunk)
|
||||
if (generateImage?.images?.length) {
|
||||
onChunk({ type: ChunkType.IMAGE_COMPLETE, image: generateImage })
|
||||
}
|
||||
if (chunk.candidates?.[0]?.finishReason) {
|
||||
if (chunk.text) {
|
||||
onChunk({ type: ChunkType.TEXT_COMPLETE, text: content })
|
||||
}
|
||||
if (chunk.usageMetadata) {
|
||||
finalUsage.prompt_tokens = chunk.usageMetadata.promptTokenCount || 0
|
||||
finalUsage.completion_tokens = chunk.usageMetadata.candidatesTokenCount || 0
|
||||
finalUsage.total_tokens = chunk.usageMetadata.totalTokenCount || 0
|
||||
}
|
||||
}
|
||||
}
|
||||
onChunk({
|
||||
type: ChunkType.BLOCK_COMPLETE,
|
||||
response: {
|
||||
usage: finalUsage,
|
||||
metrics: {
|
||||
completion_tokens: finalUsage.completion_tokens,
|
||||
time_completion_millsec: new Date().getTime() - start_time_millsec,
|
||||
time_first_token_millsec: time_first_token_millsec - start_time_millsec
|
||||
}
|
||||
}
|
||||
})
|
||||
} catch (error) {
|
||||
console.error('[generateImageByChat] error', error)
|
||||
onChunk({
|
||||
type: ChunkType.ERROR,
|
||||
error
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
public convertMcpTools<T>(mcpTools: MCPTool[]): T[] {
|
||||
|
||||
@ -34,6 +34,7 @@ import {
|
||||
MCPCallToolResponse,
|
||||
MCPTool,
|
||||
MCPToolResponse,
|
||||
Metrics,
|
||||
Model,
|
||||
Provider,
|
||||
Suggestion,
|
||||
@ -395,7 +396,6 @@ export default class OpenAICompatibleProvider extends BaseOpenAiProvider {
|
||||
return streamOutput
|
||||
}
|
||||
|
||||
const start_time_millsec = new Date().getTime()
|
||||
const lastUserMessage = _messages.findLast((m) => m.role === 'user')
|
||||
const { abortController, cleanup, signalPromise } = this.createAbortController(lastUserMessage?.id, true)
|
||||
const { signal } = abortController
|
||||
@ -423,6 +423,18 @@ export default class OpenAICompatibleProvider extends BaseOpenAiProvider {
|
||||
reqMessages = [systemMessage, ...userMessages].filter(Boolean) as ChatCompletionMessageParam[]
|
||||
}
|
||||
|
||||
let finalUsage: Usage = {
|
||||
completion_tokens: 0,
|
||||
prompt_tokens: 0,
|
||||
total_tokens: 0
|
||||
}
|
||||
|
||||
const finalMetrics: Metrics = {
|
||||
completion_tokens: 0,
|
||||
time_completion_millsec: 0,
|
||||
time_first_token_millsec: 0
|
||||
}
|
||||
|
||||
const toolResponses: MCPToolResponse[] = []
|
||||
|
||||
const processToolResults = async (toolResults: Awaited<ReturnType<typeof parseAndCallTools>>, idx: number) => {
|
||||
@ -505,18 +517,17 @@ export default class OpenAICompatibleProvider extends BaseOpenAiProvider {
|
||||
|
||||
const processStream = async (stream: any, idx: number) => {
|
||||
const toolCalls: ChatCompletionMessageToolCall[] = []
|
||||
let time_first_token_millsec = 0
|
||||
const start_time_millsec = new Date().getTime()
|
||||
|
||||
// Handle non-streaming case (already returns early, no change needed here)
|
||||
if (!isSupportStreamOutput()) {
|
||||
const time_completion_millsec = new Date().getTime() - start_time_millsec
|
||||
// Calculate final metrics once
|
||||
const finalMetrics = {
|
||||
completion_tokens: stream.usage?.completion_tokens,
|
||||
time_completion_millsec,
|
||||
time_first_token_millsec: 0 // Non-streaming, first token time is not relevant
|
||||
}
|
||||
finalMetrics.completion_tokens = stream.usage?.completion_tokens
|
||||
finalMetrics.time_completion_millsec = new Date().getTime() - start_time_millsec
|
||||
|
||||
// Create a synthetic usage object if stream.usage is undefined
|
||||
const finalUsage = stream.usage
|
||||
finalUsage = { ...stream.usage }
|
||||
// Separate onChunk calls for text and usage/metrics
|
||||
let content = ''
|
||||
stream.choices.forEach((choice) => {
|
||||
@ -526,7 +537,7 @@ export default class OpenAICompatibleProvider extends BaseOpenAiProvider {
|
||||
onChunk({
|
||||
type: ChunkType.THINKING_COMPLETE,
|
||||
text: choice.message.reasoning,
|
||||
thinking_millsec: time_completion_millsec
|
||||
thinking_millsec: new Date().getTime() - start_time_millsec
|
||||
})
|
||||
}
|
||||
// text
|
||||
@ -576,20 +587,9 @@ export default class OpenAICompatibleProvider extends BaseOpenAiProvider {
|
||||
return
|
||||
}
|
||||
|
||||
let content = '' // Accumulate content for tool processing if needed
|
||||
let content = ''
|
||||
let thinkingContent = ''
|
||||
// 记录最终的完成时间差
|
||||
let final_time_completion_millsec_delta = 0
|
||||
let final_time_thinking_millsec_delta = 0
|
||||
// Variable to store the last received usage object
|
||||
let lastUsage: Usage | undefined = undefined
|
||||
// let isThinkingInContent: ThoughtProcessor | undefined = undefined
|
||||
// const processThinkingChunk = this.handleThinkingTags()
|
||||
let isFirstChunk = true
|
||||
let time_first_token_millsec = 0
|
||||
let time_first_token_millsec_delta = 0
|
||||
let time_first_content_millsec = 0
|
||||
let time_thinking_start = 0
|
||||
|
||||
// 1. 初始化中间件
|
||||
const reasoningTags = [
|
||||
@ -640,25 +640,24 @@ export default class OpenAICompatibleProvider extends BaseOpenAiProvider {
|
||||
|
||||
// 3. 消费 processedStream，分发 onChunk
|
||||
for await (const chunk of readableStreamAsyncIterable(processedStream)) {
|
||||
const currentTime = new Date().getTime()
|
||||
const delta = chunk.type === 'finish' ? chunk.delta : chunk
|
||||
const rawChunk = chunk.type === 'finish' ? chunk.chunk : chunk
|
||||
|
||||
switch (chunk.type) {
|
||||
case 'reasoning': {
|
||||
if (time_thinking_start === 0) {
|
||||
time_thinking_start = currentTime
|
||||
time_first_token_millsec = currentTime
|
||||
time_first_token_millsec_delta = currentTime - start_time_millsec
|
||||
if (time_first_token_millsec === 0) {
|
||||
time_first_token_millsec = new Date().getTime()
|
||||
}
|
||||
thinkingContent += chunk.textDelta
|
||||
const thinking_time = currentTime - time_thinking_start
|
||||
onChunk({ type: ChunkType.THINKING_DELTA, text: chunk.textDelta, thinking_millsec: thinking_time })
|
||||
onChunk({
|
||||
type: ChunkType.THINKING_DELTA,
|
||||
text: chunk.textDelta,
|
||||
thinking_millsec: new Date().getTime() - time_first_token_millsec
|
||||
})
|
||||
break
|
||||
}
|
||||
case 'text-delta': {
|
||||
let textDelta = chunk.textDelta
|
||||
|
||||
if (assistant.enableWebSearch && delta) {
|
||||
const originalDelta = rawChunk?.choices?.[0]?.delta
|
||||
|
||||
@ -676,25 +675,32 @@ export default class OpenAICompatibleProvider extends BaseOpenAiProvider {
|
||||
if (isFirstChunk) {
|
||||
isFirstChunk = false
|
||||
if (time_first_token_millsec === 0) {
|
||||
time_first_token_millsec = currentTime
|
||||
time_first_token_millsec_delta = currentTime - start_time_millsec
|
||||
time_first_token_millsec = new Date().getTime()
|
||||
} else {
|
||||
onChunk({
|
||||
type: ChunkType.THINKING_COMPLETE,
|
||||
text: thinkingContent,
|
||||
thinking_millsec: new Date().getTime() - time_first_token_millsec
|
||||
})
|
||||
}
|
||||
}
|
||||
content += textDelta
|
||||
if (time_thinking_start > 0 && time_first_content_millsec === 0) {
|
||||
time_first_content_millsec = currentTime
|
||||
final_time_thinking_millsec_delta = time_first_content_millsec - time_thinking_start
|
||||
|
||||
onChunk({
|
||||
type: ChunkType.THINKING_COMPLETE,
|
||||
text: thinkingContent,
|
||||
thinking_millsec: final_time_thinking_millsec_delta
|
||||
})
|
||||
}
|
||||
onChunk({ type: ChunkType.TEXT_DELTA, text: textDelta })
|
||||
break
|
||||
}
|
||||
case 'tool-calls': {
|
||||
if (isFirstChunk) {
|
||||
isFirstChunk = false
|
||||
if (time_first_token_millsec === 0) {
|
||||
time_first_token_millsec = new Date().getTime()
|
||||
} else {
|
||||
onChunk({
|
||||
type: ChunkType.THINKING_COMPLETE,
|
||||
text: thinkingContent,
|
||||
thinking_millsec: new Date().getTime() - time_first_token_millsec
|
||||
})
|
||||
}
|
||||
}
|
||||
chunk.delta.tool_calls.forEach((toolCall) => {
|
||||
const { id, index, type, function: fun } = toolCall
|
||||
if (id && type === 'function' && fun) {
|
||||
@ -721,10 +727,14 @@ export default class OpenAICompatibleProvider extends BaseOpenAiProvider {
|
||||
|
||||
if (!isEmpty(finishReason)) {
|
||||
onChunk({ type: ChunkType.TEXT_COMPLETE, text: content })
|
||||
final_time_completion_millsec_delta = currentTime - start_time_millsec
|
||||
if (usage) {
|
||||
lastUsage = usage
|
||||
finalUsage.completion_tokens += usage.completion_tokens || 0
|
||||
finalUsage.prompt_tokens += usage.prompt_tokens || 0
|
||||
finalUsage.total_tokens += usage.total_tokens || 0
|
||||
finalMetrics.completion_tokens += usage.completion_tokens || 0
|
||||
}
|
||||
finalMetrics.time_completion_millsec += new Date().getTime() - start_time_millsec
|
||||
finalMetrics.time_first_token_millsec = time_first_token_millsec - start_time_millsec
|
||||
if (originalFinishDelta?.annotations) {
|
||||
onChunk({
|
||||
type: ChunkType.LLM_WEB_SEARCH_COMPLETE,
|
||||
@ -774,49 +784,46 @@ export default class OpenAICompatibleProvider extends BaseOpenAiProvider {
|
||||
} as LLMWebSearchCompleteChunk)
|
||||
}
|
||||
}
|
||||
reqMessages.push({
|
||||
role: 'assistant',
|
||||
content: content,
|
||||
tool_calls: toolCalls.length
|
||||
? toolCalls.map((toolCall) => ({
|
||||
id: toolCall.id,
|
||||
function: {
|
||||
...toolCall.function,
|
||||
arguments:
|
||||
typeof toolCall.function.arguments === 'string'
|
||||
? toolCall.function.arguments
|
||||
: JSON.stringify(toolCall.function.arguments)
|
||||
},
|
||||
type: 'function'
|
||||
}))
|
||||
: undefined
|
||||
})
|
||||
let toolResults: Awaited<ReturnType<typeof parseAndCallTools>> = []
|
||||
if (toolCalls.length) {
|
||||
toolResults = await processToolCalls(mcpTools, toolCalls)
|
||||
}
|
||||
if (content.length) {
|
||||
toolResults = toolResults.concat(await processToolUses(content))
|
||||
}
|
||||
if (toolResults.length) {
|
||||
await processToolResults(toolResults, idx)
|
||||
}
|
||||
onChunk({
|
||||
type: ChunkType.BLOCK_COMPLETE,
|
||||
response: {
|
||||
usage: lastUsage,
|
||||
metrics: {
|
||||
completion_tokens: lastUsage?.completion_tokens,
|
||||
time_completion_millsec: final_time_completion_millsec_delta,
|
||||
time_first_token_millsec: time_first_token_millsec_delta,
|
||||
time_thinking_millsec: final_time_thinking_millsec_delta
|
||||
}
|
||||
}
|
||||
})
|
||||
break
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
reqMessages.push({
|
||||
role: 'assistant',
|
||||
content: content,
|
||||
tool_calls: toolCalls.length
|
||||
? toolCalls.map((toolCall) => ({
|
||||
id: toolCall.id,
|
||||
function: {
|
||||
...toolCall.function,
|
||||
arguments:
|
||||
typeof toolCall.function.arguments === 'string'
|
||||
? toolCall.function.arguments
|
||||
: JSON.stringify(toolCall.function.arguments)
|
||||
},
|
||||
type: 'function'
|
||||
}))
|
||||
: undefined
|
||||
})
|
||||
let toolResults: Awaited<ReturnType<typeof parseAndCallTools>> = []
|
||||
if (toolCalls.length) {
|
||||
toolResults = await processToolCalls(mcpTools, toolCalls)
|
||||
}
|
||||
if (content.length) {
|
||||
toolResults = toolResults.concat(await processToolUses(content))
|
||||
}
|
||||
if (toolResults.length) {
|
||||
await processToolResults(toolResults, idx)
|
||||
}
|
||||
|
||||
onChunk({
|
||||
type: ChunkType.BLOCK_COMPLETE,
|
||||
response: {
|
||||
usage: finalUsage,
|
||||
metrics: finalMetrics
|
||||
}
|
||||
})
|
||||
}
|
||||
|
||||
reqMessages = processReqMessages(model, reqMessages)
|
||||
|
||||
@ -24,6 +24,7 @@ import {
|
||||
MCPCallToolResponse,
|
||||
MCPTool,
|
||||
MCPToolResponse,
|
||||
Metrics,
|
||||
Model,
|
||||
Provider,
|
||||
Suggestion,
|
||||
@ -332,7 +333,6 @@ export abstract class BaseOpenAiProvider extends BaseProvider {
|
||||
const lastUserMessage = _messages.findLast((m) => m.role === 'user')
|
||||
const { abortController, cleanup, signalPromise } = this.createAbortController(lastUserMessage?.id, true)
|
||||
const { signal } = abortController
|
||||
let time_first_token_millsec_delta = 0
|
||||
const start_time_millsec = new Date().getTime()
|
||||
const response = await this.sdk.chat.completions
|
||||
// @ts-ignore key is not typed
|
||||
@ -354,8 +354,17 @@ export abstract class BaseOpenAiProvider extends BaseProvider {
|
||||
const processStream = async (stream: any) => {
|
||||
let content = ''
|
||||
let isFirstChunk = true
|
||||
let final_time_completion_millsec_delta = 0
|
||||
let lastUsage: Usage | undefined = undefined
|
||||
const finalUsage: Usage = {
|
||||
completion_tokens: 0,
|
||||
prompt_tokens: 0,
|
||||
total_tokens: 0
|
||||
}
|
||||
|
||||
const finalMetrics: Metrics = {
|
||||
completion_tokens: 0,
|
||||
time_completion_millsec: 0,
|
||||
time_first_token_millsec: 0
|
||||
}
|
||||
for await (const chunk of stream as any) {
|
||||
if (window.keyv.get(EVENT_NAMES.CHAT_COMPLETION_PAUSED)) {
|
||||
break
|
||||
@ -368,17 +377,21 @@ export abstract class BaseOpenAiProvider extends BaseProvider {
|
||||
}
|
||||
if (isFirstChunk) {
|
||||
isFirstChunk = false
|
||||
time_first_token_millsec_delta = new Date().getTime() - start_time_millsec
|
||||
finalMetrics.time_first_token_millsec = new Date().getTime() - start_time_millsec
|
||||
}
|
||||
content += delta.content
|
||||
onChunk({ type: ChunkType.TEXT_DELTA, text: delta.content })
|
||||
}
|
||||
if (!isEmpty(finishReason) || chunk?.annotations) {
|
||||
onChunk({ type: ChunkType.TEXT_COMPLETE, text: content })
|
||||
final_time_completion_millsec_delta = new Date().getTime() - start_time_millsec
|
||||
finalMetrics.time_completion_millsec = new Date().getTime() - start_time_millsec
|
||||
if (chunk.usage) {
|
||||
lastUsage = chunk.usage
|
||||
const usage = chunk.usage as OpenAI.Completions.CompletionUsage
|
||||
finalUsage.completion_tokens = usage.completion_tokens
|
||||
finalUsage.prompt_tokens = usage.prompt_tokens
|
||||
finalUsage.total_tokens = usage.total_tokens
|
||||
}
|
||||
finalMetrics.completion_tokens = finalUsage.completion_tokens
|
||||
}
|
||||
if (delta?.annotations) {
|
||||
onChunk({
|
||||
@ -393,12 +406,8 @@ export abstract class BaseOpenAiProvider extends BaseProvider {
|
||||
onChunk({
|
||||
type: ChunkType.BLOCK_COMPLETE,
|
||||
response: {
|
||||
usage: lastUsage,
|
||||
metrics: {
|
||||
completion_tokens: lastUsage?.completion_tokens,
|
||||
time_completion_millsec: final_time_completion_millsec_delta,
|
||||
time_first_token_millsec: time_first_token_millsec_delta
|
||||
}
|
||||
usage: finalUsage,
|
||||
metrics: finalMetrics
|
||||
}
|
||||
})
|
||||
}
|
||||
@ -428,7 +437,6 @@ export abstract class BaseOpenAiProvider extends BaseProvider {
|
||||
type: 'input_text'
|
||||
}
|
||||
if (isSupportedReasoningEffortOpenAIModel(model)) {
|
||||
systemMessageInput.text = `Formatting re-enabled${systemMessageInput.text ? '\n' + systemMessageInput.text : ''}`
|
||||
systemMessage.role = 'developer'
|
||||
}
|
||||
|
||||
@ -455,9 +463,6 @@ export abstract class BaseOpenAiProvider extends BaseProvider {
|
||||
userMessage.push(await this.getReponseMessageParam(message, model))
|
||||
}
|
||||
|
||||
let time_first_token_millsec = 0
|
||||
const start_time_millsec = new Date().getTime()
|
||||
|
||||
const lastUserMessage = _messages.findLast((m) => m.role === 'user')
|
||||
const { abortController, cleanup, signalPromise } = this.createAbortController(lastUserMessage?.id, true)
|
||||
const { signal } = abortController
|
||||
@ -470,6 +475,18 @@ export abstract class BaseOpenAiProvider extends BaseProvider {
|
||||
reqMessages = [systemMessage, ...userMessage].filter(Boolean) as OpenAI.Responses.EasyInputMessage[]
|
||||
}
|
||||
|
||||
const finalUsage: Usage = {
|
||||
completion_tokens: 0,
|
||||
prompt_tokens: 0,
|
||||
total_tokens: 0
|
||||
}
|
||||
|
||||
const finalMetrics: Metrics = {
|
||||
completion_tokens: 0,
|
||||
time_completion_millsec: 0,
|
||||
time_first_token_millsec: 0
|
||||
}
|
||||
|
||||
const toolResponses: MCPToolResponse[] = []
|
||||
|
||||
const processToolResults = async (toolResults: Awaited<ReturnType<typeof parseAndCallTools>>, idx: number) => {
|
||||
@ -549,6 +566,8 @@ export abstract class BaseOpenAiProvider extends BaseProvider {
|
||||
idx: number
|
||||
) => {
|
||||
const toolCalls: OpenAI.Responses.ResponseFunctionToolCall[] = []
|
||||
let time_first_token_millsec = 0
|
||||
const start_time_millsec = new Date().getTime()
|
||||
|
||||
if (!streamOutput) {
|
||||
const nonStream = stream as OpenAI.Responses.Response
|
||||
@ -633,17 +652,15 @@ export abstract class BaseOpenAiProvider extends BaseProvider {
|
||||
|
||||
const outputItems: OpenAI.Responses.ResponseOutputItem[] = []
|
||||
|
||||
let lastUsage: Usage | undefined = undefined
|
||||
let final_time_completion_millsec_delta = 0
|
||||
for await (const chunk of stream as Stream<OpenAI.Responses.ResponseStreamEvent>) {
|
||||
if (window.keyv.get(EVENT_NAMES.CHAT_COMPLETION_PAUSED)) {
|
||||
break
|
||||
}
|
||||
switch (chunk.type) {
|
||||
case 'response.created':
|
||||
time_first_token_millsec = new Date().getTime()
|
||||
break
|
||||
case 'response.output_item.added':
|
||||
if (time_first_token_millsec === 0) {
|
||||
time_first_token_millsec = new Date().getTime()
|
||||
}
|
||||
if (chunk.item.type === 'function_call') {
|
||||
outputItems.push(chunk.item)
|
||||
}
|
||||
@ -708,18 +725,18 @@ export abstract class BaseOpenAiProvider extends BaseProvider {
|
||||
}
|
||||
break
|
||||
case 'response.completed': {
|
||||
final_time_completion_millsec_delta = new Date().getTime() - start_time_millsec
|
||||
const completion_tokens =
|
||||
(chunk.response.usage?.output_tokens || 0) +
|
||||
(chunk.response.usage?.output_tokens_details.reasoning_tokens ?? 0)
|
||||
const total_tokens =
|
||||
(chunk.response.usage?.total_tokens || 0) +
|
||||
(chunk.response.usage?.output_tokens_details.reasoning_tokens ?? 0)
|
||||
lastUsage = {
|
||||
completion_tokens,
|
||||
prompt_tokens: chunk.response.usage?.input_tokens || 0,
|
||||
total_tokens
|
||||
}
|
||||
finalUsage.completion_tokens += completion_tokens
|
||||
finalUsage.prompt_tokens += chunk.response.usage?.input_tokens || 0
|
||||
finalUsage.total_tokens += total_tokens
|
||||
finalMetrics.completion_tokens += completion_tokens
|
||||
finalMetrics.time_completion_millsec += new Date().getTime() - start_time_millsec
|
||||
finalMetrics.time_first_token_millsec = time_first_token_millsec - start_time_millsec
|
||||
break
|
||||
}
|
||||
case 'error':
|
||||
@ -761,12 +778,8 @@ export abstract class BaseOpenAiProvider extends BaseProvider {
|
||||
onChunk({
|
||||
type: ChunkType.BLOCK_COMPLETE,
|
||||
response: {
|
||||
usage: lastUsage,
|
||||
metrics: {
|
||||
completion_tokens: lastUsage?.completion_tokens,
|
||||
time_completion_millsec: final_time_completion_millsec_delta,
|
||||
time_first_token_millsec: time_first_token_millsec - start_time_millsec
|
||||
}
|
||||
usage: finalUsage,
|
||||
metrics: finalMetrics
|
||||
}
|
||||
})
|
||||
}
|
||||
|
||||
@ -59,7 +59,7 @@ export function createStreamProcessor(callbacks: StreamProcessorCallbacks = {})
|
||||
callbacks.onTextComplete(data.text)
|
||||
}
|
||||
if (data.type === ChunkType.THINKING_DELTA && callbacks.onThinkingChunk) {
|
||||
callbacks.onThinkingChunk(data.text, data.thinking_millsec)
|
||||
callbacks.onThinkingChunk(data.text)
|
||||
}
|
||||
if (data.type === ChunkType.THINKING_COMPLETE && callbacks.onThinkingComplete) {
|
||||
callbacks.onThinkingComplete(data.text, data.thinking_millsec)
|
||||
|
||||
@ -46,7 +46,7 @@ const persistedReducer = persistReducer(
|
||||
{
|
||||
key: 'cherry-studio',
|
||||
storage,
|
||||
version: 99,
|
||||
version: 100,
|
||||
blacklist: ['runtime', 'messages', 'messageBlocks'],
|
||||
migrate
|
||||
},
|
||||
|
||||
@ -1296,6 +1296,16 @@ const migrateConfig = {
|
||||
} catch (error) {
|
||||
return state
|
||||
}
|
||||
},
|
||||
'100': (state: RootState) => {
|
||||
try {
|
||||
if (!state.settings.zoomFactor) {
|
||||
state.settings.zoomFactor = 1
|
||||
}
|
||||
return state
|
||||
} catch (error) {
|
||||
return state
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@ -398,7 +398,7 @@ const fetchAndProcessAssistantResponseImpl = async (
|
||||
} else {
|
||||
const newBlock = createThinkingBlock(assistantMsgId, accumulatedThinking, {
|
||||
status: MessageBlockStatus.STREAMING,
|
||||
thinking_millsec: thinking_millsec
|
||||
thinking_millsec: 0
|
||||
})
|
||||
handleBlockTransition(newBlock, MessageBlockType.THINKING)
|
||||
}
|
||||
@ -565,7 +565,7 @@ const fetchAndProcessAssistantResponseImpl = async (
|
||||
message: pauseErrorLanguagePlaceholder || error.message || 'Stream processing error',
|
||||
originalMessage: error.message,
|
||||
stack: error.stack,
|
||||
status: error.status,
|
||||
status: error.status || error.code,
|
||||
requestId: error.request_id
|
||||
}
|
||||
if (lastBlockId) {
|
||||
@ -609,13 +609,13 @@ const fetchAndProcessAssistantResponseImpl = async (
|
||||
// 更新topic的name
|
||||
autoRenameTopic(assistant, topicId)
|
||||
|
||||
if (response && !response.usage) {
|
||||
if (response && response.usage?.total_tokens === 0) {
|
||||
const usage = await estimateMessagesUsage({ assistant, messages: finalContextWithAssistant })
|
||||
response.usage = usage
|
||||
}
|
||||
}
|
||||
if (response && response.metrics) {
|
||||
if (!response.metrics.completion_tokens && response.usage) {
|
||||
if (response.metrics.completion_tokens === 0 && response.usage?.completion_tokens) {
|
||||
response = {
|
||||
...response,
|
||||
metrics: {
|
||||
|
||||
@ -121,8 +121,8 @@ export type Usage = OpenAI.Completions.CompletionUsage & {
|
||||
}
|
||||
|
||||
export type Metrics = {
|
||||
completion_tokens?: number
|
||||
time_completion_millsec?: number
|
||||
completion_tokens: number
|
||||
time_completion_millsec: number
|
||||
time_first_token_millsec?: number
|
||||
time_thinking_millsec?: number
|
||||
}
|
||||
|
||||
Loading…
Reference in New Issue
Block a user