refactor: 推理请求统一走 er_backend 代理（127.0.0.1:17700）
- 嵌入/生成服务地址不再由前端按 cardType 推导端口，改为经 er_backend 的 /v1/* 代理转发，服务端口以后端配置为唯一真相来源 - config.ts: 新增 ER_BACKEND_URL 统一常量，删除端口推导逻辑 - generator/statusHandler/queryHandler/erBackendHandler 跟随改造 - fix: package-kylin-arm64-deb.sh 从 .app-version 读版本号时剥离 v 前缀， 修复无 git 环境下 dpkg-deb 报"版本号没有以数字开头"
This commit is contained in:
surui
parent
a319caeff6
commit
836d25cad2
@ -29,7 +29,7 @@ fi
|
||||
# 的解析逻辑保持一致，确保三处版本号（deb 文件名、界面显示、控制文件）永远一致。
|
||||
VERSION=$(git describe --tags --abbrev=0 2>/dev/null | sed 's/^v//') || true
|
||||
if [ -z "$VERSION" ] && [ -f .app-version ]; then
|
||||
VERSION=$(cat .app-version | tr -d '[:space:]')
|
||||
VERSION=$(cat .app-version | tr -d '[:space:]' | sed 's/^v//')
|
||||
fi
|
||||
if [ -z "$VERSION" ]; then
|
||||
VERSION=$(node -p "require('./package.json').version")
|
||||
|
||||
@ -2,11 +2,14 @@ import { app } from 'electron'
|
||||
import { join } from 'path'
|
||||
import { existsSync, readFileSync, writeFileSync } from 'fs'
|
||||
|
||||
// 推理请求统一走 er_backend 代理（管理面端口，/v1/* 由后端转发到其实际管理的 llama-server）
|
||||
export const ER_BACKEND_URL = 'http://127.0.0.1:17700'
|
||||
|
||||
export interface AppConfig {
|
||||
embed: {
|
||||
provider: 'local' | 'llamacpp'
|
||||
localModelPath: string
|
||||
baseUrl: string // 由 applyPlatformOverrides 自动推导，不对用户暴露
|
||||
baseUrl: string // 由 applyPlatformOverrides 固定为 er_backend 代理地址，不对用户暴露
|
||||
}
|
||||
rag: {
|
||||
chunkSize: number
|
||||
@ -14,6 +17,7 @@ export interface AppConfig {
|
||||
topK: number
|
||||
systemPrompt: string
|
||||
}
|
||||
// 服务的启动参数/端口由 er_backend 统一管理，以下字段仅为历史配置保留，前端不再用于路由
|
||||
llamacppServer: {
|
||||
executablePath: string
|
||||
modelPath: string
|
||||
@ -60,7 +64,7 @@ export const DEFAULT_CONFIG: AppConfig = {
|
||||
embed: {
|
||||
provider: 'llamacpp',
|
||||
localModelPath: '',
|
||||
baseUrl: 'http://127.0.0.1:17701'
|
||||
baseUrl: ER_BACKEND_URL
|
||||
},
|
||||
llamacppServer: {
|
||||
executablePath: '',
|
||||
@ -206,19 +210,18 @@ export function getConfig(): AppConfig {
|
||||
}
|
||||
|
||||
// 按计算卡型号覆盖 embed provider 和默认启动参数
|
||||
// 推理请求统一走 er_backend 代理（ER_BACKEND_URL），服务端口以后端配置为准，前端不再维护
|
||||
function applyPlatformOverrides(config: AppConfig): void {
|
||||
const card = config.hardware.cardType
|
||||
if (card === 'er25') {
|
||||
// ER25：嵌入固定本地 CPU，无嵌入服务进程
|
||||
config.embed.provider = 'local'
|
||||
config.llamacppServer.port = 8080
|
||||
if (!config.llamacppServer.extraArgs.trim()) {
|
||||
config.llamacppServer.extraArgs = '--port 8080 --ctk bf16 --ctv bf16 --no-warmup -ub 128'
|
||||
}
|
||||
} else {
|
||||
// ER160：嵌入默认 llamacpp NPU（local 保持不变）
|
||||
if (config.embed.provider !== 'local') config.embed.provider = 'llamacpp'
|
||||
config.llamacppServer.port = 8081
|
||||
if (!config.llamacppServer.extraArgs.trim()) {
|
||||
config.llamacppServer.extraArgs = '--port 8081 --presence_penalty 1.5'
|
||||
}
|
||||
@ -226,8 +229,8 @@ function applyPlatformOverrides(config: AppConfig): void {
|
||||
config.llamacppEmbed.extraArgs = DEFAULT_CONFIG.llamacppEmbed.extraArgs
|
||||
}
|
||||
}
|
||||
// 嵌入服务地址跟随端口
|
||||
config.embed.baseUrl = `http://127.0.0.1:${config.llamacppEmbed.port}`
|
||||
// 嵌入/生成服务地址固定为 er_backend 代理
|
||||
config.embed.baseUrl = ER_BACKEND_URL
|
||||
}
|
||||
|
||||
export function saveConfig(config: AppConfig): void {
|
||||
|
||||
@ -1,7 +1,6 @@
|
||||
import { ipcMain } from 'electron'
|
||||
import { execFile } from 'child_process'
|
||||
|
||||
const ER_BACKEND_URL = 'http://localhost:17700'
|
||||
import { ER_BACKEND_URL } from '../config'
|
||||
|
||||
async function isRunning(): Promise<boolean> {
|
||||
try {
|
||||
|
||||
@ -16,7 +16,7 @@ export function registerQueryHandlers(getWin: () => BrowserWindow | null): void
|
||||
console.log(`\n${'='.repeat(60)}`)
|
||||
console.log('[query] 问题：', payload.question)
|
||||
console.log('[query] topK：', payload.topK)
|
||||
console.log('[query] 向量化方式：', config.embed.provider, '生成服务端口：', config.llamacppServer.port)
|
||||
console.log('[query] 向量化方式：', config.embed.provider, '生成/嵌入服务：er_backend 代理（127.0.0.1:17700）')
|
||||
|
||||
const send = (channel: string, data: unknown): void => {
|
||||
getWin()?.webContents.send(channel, data)
|
||||
@ -50,7 +50,6 @@ export function registerQueryHandlers(getWin: () => BrowserWindow | null): void
|
||||
payload.question,
|
||||
chunks,
|
||||
payload.systemPrompt,
|
||||
config.llamacppServer,
|
||||
{
|
||||
onDelta: (delta) => send('query:delta', { delta }),
|
||||
onDone: (stats) => send('query:done', { stats: stats ?? null }),
|
||||
|
||||
@ -2,7 +2,7 @@ import { ipcMain, dialog } from 'electron'
|
||||
import type { BrowserWindow } from 'electron'
|
||||
import { existsSync, readFileSync, writeFileSync } from 'fs'
|
||||
import { join } from 'path'
|
||||
import { getConfig, saveConfig, getDbPath, AppConfig } from '../config'
|
||||
import { getConfig, saveConfig, getDbPath, AppConfig, ER_BACKEND_URL } from '../config'
|
||||
import { getDocCount } from '../db/lancedb'
|
||||
import { embedOneLocal, resolveLocalModelPath } from '../rag/localEmbedder'
|
||||
import { syncNpuPolling } from './npuHandler'
|
||||
@ -100,11 +100,12 @@ export function registerStatusHandlers(getWin: () => BrowserWindow | null): void
|
||||
}
|
||||
})
|
||||
|
||||
ipcMain.handle('chat-test', async (_event, llamacppConfig: AppConfig['llamacppServer']): Promise<TestResult> => {
|
||||
// 生成服务连接测试（经 er_backend 代理转发到 llama-server）
|
||||
ipcMain.handle('chat-test', async (_event, _llamacppConfig: AppConfig['llamacppServer']): Promise<TestResult> => {
|
||||
const start = Date.now()
|
||||
const elapsed = (): string => `${Date.now() - start}ms`
|
||||
try {
|
||||
const res = await fetch(`http://127.0.0.1:${llamacppConfig.port}/v1/chat/completions`, {
|
||||
const res = await fetch(`${ER_BACKEND_URL}/v1/chat/completions`, {
|
||||
method: 'POST',
|
||||
headers: { 'Content-Type': 'application/json' },
|
||||
body: JSON.stringify({
|
||||
|
||||
@ -1,5 +1,5 @@
|
||||
import type { SearchResult } from '../db/lancedb'
|
||||
import type { AppConfig } from '../config'
|
||||
import { ER_BACKEND_URL } from '../config'
|
||||
|
||||
export interface GeneratorStats {
|
||||
tokensPerSecond: number
|
||||
@ -14,12 +14,11 @@ export interface GeneratorCallbacks {
|
||||
onError: (message: string) => void
|
||||
}
|
||||
|
||||
// 生成服务固定使用 llamacpp（ER160/ER25 均通过 er_backend 管理的 llama-server）
|
||||
// 生成服务经 er_backend 代理访问其管理的 llama-server（OpenAI 兼容接口，SSE 流式）
|
||||
export async function streamGenerate(
|
||||
question: string,
|
||||
chunks: SearchResult[],
|
||||
systemPrompt: string,
|
||||
llamacppConfig: AppConfig['llamacppServer'],
|
||||
callbacks: GeneratorCallbacks
|
||||
): Promise<void> {
|
||||
const context = chunks
|
||||
@ -33,22 +32,21 @@ ${context}
|
||||
请回答以下问题：${question}`
|
||||
|
||||
console.log('\n[generator] ---- 发送给 LLM 的完整消息 ----')
|
||||
console.log(`[generator] 服务地址：http://127.0.0.1:${llamacppConfig.port}`)
|
||||
console.log(`[generator] 服务地址：${ER_BACKEND_URL}（er_backend 代理）`)
|
||||
console.log('[generator] system prompt：\n', systemPrompt)
|
||||
console.log('[generator] user message：\n', userMessage)
|
||||
console.log('[generator] ---- 消息结束 ----\n')
|
||||
|
||||
await streamGenerateLlamaCpp(userMessage, systemPrompt, llamacppConfig, callbacks)
|
||||
await streamGenerateLlamaCpp(userMessage, systemPrompt, callbacks)
|
||||
}
|
||||
|
||||
// llama.cpp server OpenAI 兼容接口（/v1/chat/completions），SSE 格式
|
||||
async function streamGenerateLlamaCpp(
|
||||
userMessage: string,
|
||||
systemPrompt: string,
|
||||
llamacppConfig: AppConfig['llamacppServer'],
|
||||
callbacks: GeneratorCallbacks
|
||||
): Promise<void> {
|
||||
const baseUrl = `http://127.0.0.1:${llamacppConfig.port}`
|
||||
const baseUrl = ER_BACKEND_URL
|
||||
let fullResponse = ''
|
||||
let thinkOpen = false
|
||||
const requestSentTime = Date.now()
|
||||
@ -71,12 +69,12 @@ async function streamGenerateLlamaCpp(
|
||||
})
|
||||
})
|
||||
} catch (e) {
|
||||
callbacks.onError(`无法连接 llama.cpp server（${baseUrl}）：${String(e)}`)
|
||||
callbacks.onError(`无法连接生成服务（${baseUrl}，er_backend 代理）：${String(e)}`)
|
||||
return
|
||||
}
|
||||
|
||||
if (!response.ok) {
|
||||
callbacks.onError(`llama.cpp server 请求失败：${response.status}`)
|
||||
callbacks.onError(`生成服务请求失败：${response.status}`)
|
||||
return
|
||||
}
|
||||
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user