From ad91729533e2b0bc7dd5d63689c35165d75aed18 Mon Sep 17 00:00:00 2001 From: adolphzhang Date: Mon, 5 Aug 2024 00:01:24 +0800 Subject: [PATCH] pref: adjust function call position --- adapter/docker/docker-compose.yml | 2 +- adapter/vercel/utils/duckduckgo.js | 6 +- dist/buildinfo.json | 2 +- dist/index.js | 251 ++++++++++++++++++----------- dist/timestamp | 2 +- src/agent/llm.js | 141 ++-------------- src/agent/openai.js | 27 ++-- src/agent/request.js | 7 +- src/agent/toolHander.js | 127 +++++++++++++++ src/config/env.js | 3 +- src/config/middle.js | 7 +- src/prompt/tools.js | 12 ++ src/telegram/message.js | 118 -------------- src/tools/duckduckgo.js | 12 +- src/tools/jina.js | 12 +- src/utils/utils.js | 10 +- 16 files changed, 346 insertions(+), 393 deletions(-) create mode 100644 src/agent/toolHander.js create mode 100644 src/prompt/tools.js diff --git a/adapter/docker/docker-compose.yml b/adapter/docker/docker-compose.yml index 38e74963..645b1707 100644 --- a/adapter/docker/docker-compose.yml +++ b/adapter/docker/docker-compose.yml @@ -8,7 +8,7 @@ services: ports: - "8787:8787" volumes: - - ./config.toml:/app/config.toml + - ./config/config.toml:/app/config.toml environment: - REDIS_URL=redis://redis:6379 # 设置为空时读取请求的host作为DOMAIN diff --git a/adapter/vercel/utils/duckduckgo.js b/adapter/vercel/utils/duckduckgo.js index 8e90de0e..d2fae188 100644 --- a/adapter/vercel/utils/duckduckgo.js +++ b/adapter/vercel/utils/duckduckgo.js @@ -254,9 +254,5 @@ export const duckduckgo_search = { console.log(content); return { content, time }; }, - settings: { - after_prompt: "作为智能助手，请按照以下步骤有效分析并提取我提供的搜索结果，以简洁明了的方式回答我的问题：\n\n1. 阅读和评估：仔细阅读所有搜索结果，识别并优先获取来自可靠和最新来源的信息。考虑因素包括官方来源、知名机构以及信息的更新时间。\n\n2. 提取关键信息：\n • *汇率查询*：提供最新汇率并进行必要的换算。\n • *天气查询*：提供具体地点和时间的天气预报。\n • *事实性问题*：找出权威回答。\n\n3. 简洁回答：对提取的信息进行综合分析，给出简明扼要的回答。\n\n4. 识别不确定性：如果信息存在矛盾或不确定性，请解释可能原因。\n\n5. 说明信息不足：如果搜索结果无法完全回答问题，指出需要的额外信息。\n\n6. 用户友好：使用简单易懂的语言，必要时提供简短解释，确保回答易于理解。\n\n7. 附加信息：根据需要提供额外相关信息或建议，以增强回答的价值。\n\n8. 来源标注：在回答中清晰标注信息来源，包括来源网站或机构名称及数据的发布或更新时间。\n\n9. 参考列表：如果引用了多个来源，在回答最后提供简短的参考列表，列出主要信息来源。\n\n请确保目标是提供最新、最相关和最有用的信息，直接回应我的问题。避免冗长的细节，聚焦于我最关心的核心答案，并通过可靠的来源增强回答的可信度。", - after_render: (question, result) => `问题：${question}\n\n搜索结果：${result}`, - }, - // after_history_length: 0, + type: 'search' }; diff --git a/dist/buildinfo.json b/dist/buildinfo.json index 2aeb5706..625a5ce9 100644 --- a/dist/buildinfo.json +++ b/dist/buildinfo.json @@ -1 +1 @@ -{"sha": "0c3e541", "timestamp": 1722687419} +{"sha": "a5e987f", "timestamp": 1722787231} diff --git a/dist/index.js b/dist/index.js index 0652c718..da961a8b 100644 --- a/dist/index.js +++ b/dist/index.js @@ -174,7 +174,7 @@ var UserConfig = class { CURRENT_MODE = "default"; JINA_API_KEY = ""; // openai格式调用FUNCTION CALL参数 - FUNCTION_CALL_MODEL = ""; + FUNCTION_CALL_MODEL = "gpt-4o-mini"; FUNCTION_CALL_API_KEY = ""; FUNCTION_CALL_BASE = ""; }; @@ -182,9 +182,9 @@ var Environment = class { // -- 版本数据 -- // // 当前版本 - BUILD_TIMESTAMP = 1722687419; + BUILD_TIMESTAMP = 1722787231; // 当前版本 commit id - BUILD_VERSION = "0c3e541"; + BUILD_VERSION = "a5e987f"; // -- 基础配置 -- /** * @type {I18n | null} @@ -292,6 +292,7 @@ var Environment = class { // 需要使用的函数 当前仅 duckduckgo_search 和jina_reader // '["duckduckgo_search", "jina_reader"]' USE_TOOLS = []; + FUNC_LOOP_TIMES = 1; }; var ENV_KEY_MAPPER = { CHAT_MODEL: "OPENAI_CHAT_MODEL", @@ -690,6 +691,7 @@ function fetchWithRetryFunc() { const RETRY_MULTIPLIER = 2; const DEFAULT_RETRY_AFTER = 10; return async (url, options, retries = MAX_RETRIES, delayMs = RETRY_DELAY_MS) => { + let errorMsg = ""; while (retries > 0) { try { const parsedUrl = new URL(url); @@ -713,23 +715,24 @@ function fetchWithRetryFunc() { console.log(`[DONE] after ${MAX_RETRIES - retries} times`); return resp; } - const clone_resp = await resp.clone().json(); - console.log(`${JSON.stringify(clone_resp)}`); + const clone_resp = await resp.clone().text(); + console.error(`Error fetch: ${clone_resp}`); if (resp.status === 429) { const retryAfter = resp.headers.get("Retry-After") || DEFAULT_RETRY_AFTER; status429RetryTime[domain] = Date.now() + 1e3 * retryAfter; return resp; } else { - return resp; + throw new Error(clone_resp); } } catch (error) { + errorMsg = error.message; console.log(`Request failed, retry after ${delayMs / 1e3} s: ${error}`); } await delay(delayMs); delayMs *= RETRY_MULTIPLIER; retries--; } - throw new Error("Failed after maximum retries"); + throw new Error(`Failed after maximum retries: ${errorMsg}`); }; } var fetchWithRetry = fetchWithRetryFunc(); @@ -1306,7 +1309,8 @@ ${JSON.stringify(header)} body: ${JSON.stringify(body, null, 2)}`); } - const resp = await fetchWithRetry(url, { + context._info.updateStartTime(); + const resp = await fetch(url, { method: "POST", headers: header, body: JSON.stringify(body), @@ -1321,7 +1325,7 @@ ${JSON.stringify(body, null, 2)}`); const stream = options.streamBuilder(resp, controller); let contentFull = ""; let lengthDelta = 0; - let updateStep = 10; + let updateStep = 20; let msgPromise = null; let lastChunk = null; let usage = null; @@ -1375,6 +1379,120 @@ ERROR: ${e.message}`; } } +// src/prompt/tools.js +var tools_default = { + search: { + prompt: "\u4F5C\u4E3A\u667A\u80FD\u52A9\u624B\uFF0C\u8BF7\u6309\u7167\u4EE5\u4E0B\u6B65\u9AA4\u6709\u6548\u5206\u6790\u5E76\u63D0\u53D6\u6211\u63D0\u4F9B\u7684\u641C\u7D22\u7ED3\u679C\uFF0C\u4EE5\u7B80\u6D01\u660E\u4E86\u7684\u65B9\u5F0F\u56DE\u7B54\u6211\u7684\u95EE\u9898\uFF1A\n\n1. \u9605\u8BFB\u548C\u8BC4\u4F30\uFF1A\u4ED4\u7EC6\u9605\u8BFB\u6240\u6709\u641C\u7D22\u7ED3\u679C\uFF0C\u8BC6\u522B\u5E76\u4F18\u5148\u83B7\u53D6\u6765\u81EA\u53EF\u9760\u548C\u6700\u65B0\u6765\u6E90\u7684\u4FE1\u606F\u3002\u8003\u8651\u56E0\u7D20\u5305\u62EC\u5B98\u65B9\u6765\u6E90\u3001\u77E5\u540D\u673A\u6784\u4EE5\u53CA\u4FE1\u606F\u7684\u66F4\u65B0\u65F6\u95F4\u3002\n\n2. \u63D0\u53D6\u5173\u952E\u4FE1\u606F\uFF1A\n \u2022 *\u6C47\u7387\u67E5\u8BE2*\uFF1A\u63D0\u4F9B\u6700\u65B0\u6C47\u7387\u5E76\u8FDB\u884C\u5FC5\u8981\u7684\u6362\u7B97\u3002\n \u2022 *\u5929\u6C14\u67E5\u8BE2*\uFF1A\u63D0\u4F9B\u5177\u4F53\u5730\u70B9\u548C\u65F6\u95F4\u7684\u5929\u6C14\u9884\u62A5\u3002\n \u2022 *\u4E8B\u5B9E\u6027\u95EE\u9898*\uFF1A\u627E\u51FA\u6743\u5A01\u56DE\u7B54\u3002\n\n3. \u7B80\u6D01\u56DE\u7B54\uFF1A\u5BF9\u63D0\u53D6\u7684\u4FE1\u606F\u8FDB\u884C\u7EFC\u5408\u5206\u6790\uFF0C\u7ED9\u51FA\u7B80\u660E\u627C\u8981\u7684\u56DE\u7B54\u3002\n\n4. \u8BC6\u522B\u4E0D\u786E\u5B9A\u6027\uFF1A\u5982\u679C\u4FE1\u606F\u5B58\u5728\u77DB\u76FE\u6216\u4E0D\u786E\u5B9A\u6027\uFF0C\u8BF7\u89E3\u91CA\u53EF\u80FD\u539F\u56E0\u3002\n\n5. \u8BF4\u660E\u4FE1\u606F\u4E0D\u8DB3\uFF1A\u5982\u679C\u641C\u7D22\u7ED3\u679C\u65E0\u6CD5\u5B8C\u5168\u56DE\u7B54\u95EE\u9898\uFF0C\u6307\u51FA\u9700\u8981\u7684\u989D\u5916\u4FE1\u606F\u3002\n\n6. \u7528\u6237\u53CB\u597D\uFF1A\u4F7F\u7528\u7B80\u5355\u6613\u61C2\u7684\u8BED\u8A00\uFF0C\u5FC5\u8981\u65F6\u63D0\u4F9B\u7B80\u77ED\u89E3\u91CA\uFF0C\u786E\u4FDD\u56DE\u7B54\u6613\u4E8E\u7406\u89E3\u3002\n\n7. \u9644\u52A0\u4FE1\u606F\uFF1A\u6839\u636E\u9700\u8981\u63D0\u4F9B\u989D\u5916\u76F8\u5173\u4FE1\u606F\u6216\u5EFA\u8BAE\uFF0C\u4EE5\u589E\u5F3A\u56DE\u7B54\u7684\u4EF7\u503C\u3002\n\n8. \u6765\u6E90\u6807\u6CE8\uFF1A\u5728\u56DE\u7B54\u4E2D\u6E05\u6670\u6807\u6CE8\u4FE1\u606F\u6765\u6E90\uFF0C\u5305\u62EC\u6765\u6E90\u7F51\u7AD9\u6216\u673A\u6784\u540D\u79F0\u53CA\u6570\u636E\u7684\u53D1\u5E03\u6216\u66F4\u65B0\u65F6\u95F4\u3002\n\n9. \u53C2\u8003\u5217\u8868\uFF1A\u5982\u679C\u5F15\u7528\u4E86\u591A\u4E2A\u6765\u6E90\uFF0C\u5728\u56DE\u7B54\u6700\u540E\u63D0\u4F9B\u7B80\u77ED\u7684\u53C2\u8003\u5217\u8868\uFF0C\u5217\u51FA\u4E3B\u8981\u4FE1\u606F\u6765\u6E90\u3002\n\n\u8BF7\u786E\u4FDD\u76EE\u6807\u662F\u63D0\u4F9B\u6700\u65B0\u3001\u6700\u76F8\u5173\u548C\u6700\u6709\u7528\u7684\u4FE1\u606F\uFF0C\u76F4\u63A5\u56DE\u5E94\u6211\u7684\u95EE\u9898\u3002\u907F\u514D\u5197\u957F\u7684\u7EC6\u8282\uFF0C\u805A\u7126\u4E8E\u6211\u6700\u5173\u5FC3\u7684\u6838\u5FC3\u7B54\u6848\uFF0C\u5E76\u901A\u8FC7\u53EF\u9760\u7684\u6765\u6E90\u589E\u5F3A\u56DE\u7B54\u7684\u53EF\u4FE1\u5EA6\u3002", + render: (question, result) => `\u95EE\u9898: +${question} + +\u641C\u7D22\u7ED3\u679C: +${result}` + }, + web_crawler: { + prompt: '\u4F5C\u4E3A\u4E00\u4E2A\u9AD8\u6548\u7684\u5185\u5BB9\u5206\u6790\u548C\u603B\u7ED3\u52A9\u624B\uFF0C\u4F60\u7684\u4EFB\u52A1\u662F\u5BF9\u7528\u6237\u63D0\u4F9B\u7684\u7F51\u9875\u6216PDF\u5185\u5BB9\u8FDB\u884C\u5168\u9762\u800C\u7B80\u6D01\u7684\u603B\u7ED3\u3002\u8BF7\u9075\u5FAA\u4EE5\u4E0B\u6307\u5357\uFF1A\n 1. \u4ED4\u7EC6\u9605\u8BFB\u7528\u6237\u63D0\u4F9B\u7684\u5168\u90E8\u5185\u5BB9\uFF0C\u786E\u4FDD\u7406\u89E3\u4E3B\u8981\u89C2\u70B9\u548C\u5173\u952E\u4FE1\u606F\u3002\n 2. \u8BC6\u522B\u5E76\u63D0\u70BC\u51FA\u5185\u5BB9\u7684\u6838\u5FC3\u4E3B\u9898\u548C\u4E3B\u8981\u8BBA\u70B9\u3002\n 3. \u603B\u7ED3\u65F6\u5E94\u5305\u62EC\u4EE5\u4E0B\u8981\u7D20\uFF1A\n \u2022 \u5185\u5BB9\u7684\u4E3B\u8981\u76EE\u7684\u6216\u4E3B\u9898\n \u2022 \u5173\u952E\u89C2\u70B9\u6216\u8BBA\u636E\n \u2022 \u91CD\u8981\u7684\u6570\u636E\u6216\u7EDF\u8BA1\u4FE1\u606F\uFF08\u5982\u679C\u6709\uFF09\n \u2022 \u4F5C\u8005\u7684\u7ED3\u8BBA\u6216\u5EFA\u8BAE\uFF08\u5982\u679C\u9002\u7528\uFF09\n 4. \u4FDD\u6301\u5BA2\u89C2\u6027\uFF0C\u51C6\u786E\u53CD\u6620\u539F\u6587\u7684\u89C2\u70B9\uFF0C\u4E0D\u6DFB\u52A0\u4E2A\u4EBA\u89E3\u91CA\u6216\u8BC4\u8BBA\u3002\n 5. \u4F7F\u7528\u6E05\u6670\u3001\u7B80\u6D01\u7684\u8BED\u8A00\uFF0C\u907F\u514D\u4F7F\u7528\u8FC7\u4E8E\u4E13\u4E1A\u6216\u6666\u6DA9\u7684\u672F\u8BED\u3002\n 6. \u603B\u7ED3\u7684\u957F\u5EA6\u5E94\u8BE5\u662F\u539F\u6587\u768410-15%\uFF0C\u9664\u975E\u7528\u6237\u7279\u522B\u6307\u5B9A\u5176\u4ED6\u957F\u5EA6\u8981\u6C42\u3002\n 7. \u5982\u679C\u5185\u5BB9\u5305\u542B\u591A\u4E2A\u90E8\u5206\u6216\u7AE0\u8282\uFF0C\u53EF\u4EE5\u4F7F\u7528\u7B80\u77ED\u7684\u5C0F\u6807\u9898\u6765\u7EC4\u7EC7\u4F60\u7684\u603B\u7ED3\u3002\n 8. \u5982\u679C\u539F\u6587\u5305\u542B\u56FE\u8868\u6216\u56FE\u50CF\u7684\u91CD\u8981\u4FE1\u606F\uFF0C\u8BF7\u5728\u603B\u7ED3\u4E2D\u63D0\u53CA\u8FD9\u4E00\u70B9\u3002\n 9. \u5982\u679C\u5185\u5BB9\u6D89\u53CA\u65F6\u95F4\u654F\u611F\u7684\u4FE1\u606F\uFF0C\u8BF7\u5728\u603B\u7ED3\u4E2D\u6CE8\u660E\u5185\u5BB9\u7684\u53D1\u5E03\u65E5\u671F\u6216\u7248\u672C\u3002\n 10. \u5982\u679C\u539F\u6587\u5B58\u5728\u660E\u663E\u7684\u504F\u89C1\u6216\u4E89\u8BAE\u6027\u89C2\u70B9\uFF0C\u8BF7\u5728\u603B\u7ED3\u4E2D\u5BA2\u89C2\u5730\u6307\u51FA\u8FD9\u4E00\u70B9\u3002\n 11. \u603B\u7ED3\u5B8C\u6210\u540E\uFF0C\u63D0\u4F9B1-3\u4E2A\u5173\u952E\u8BCD\u6216\u77ED\u8BED\uFF0C\u6982\u62EC\u5185\u5BB9\u7684\u6838\u5FC3\u4E3B\u9898\u3002\n 12. \u5982\u679C\u7528\u6237\u8981\u6C42\uFF0C\u53EF\u4EE5\u5728\u603B\u7ED3\u7684\u6700\u540E\u6DFB\u52A0\u4E00\u4E2A\u7B80\u77ED\u7684"\u8FDB\u4E00\u6B65\u9605\u8BFB\u5EFA\u8BAE"\u90E8\u5206\u3002\n \u8BF7\u8BB0\u4F4F\uFF0C\u4F60\u7684\u76EE\u6807\u662F\u63D0\u4F9B\u4E00\u4E2A\u5168\u9762\u3001\u51C6\u786E\u3001\u6613\u4E8E\u7406\u89E3\u7684\u603B\u7ED3\uFF0C\u5E2E\u52A9\u7528\u6237\u5FEB\u901F\u628A\u63E1\u5185\u5BB9\u7684\u7CBE\u9AD3\u3002\u5982\u679C\u5185\u5BB9\u7279\u522B\u957F\u6216\u590D\u6742\uFF0C\u4F60\u53EF\u4EE5\u8BE2\u95EE\u7528\u6237\u662F\u5426\u9700\u8981\u66F4\u8BE6\u7EC6\u7684\u603B\u7ED3\u6216\u7279\u5B9A\u90E8\u5206\u7684\u6DF1\u5165\u5206\u6790\u3002', + render: (question, result) => `${question && question + "\n"}\u7F51\u9875\u5185\u5BB9: +${result}` + } +}; + +// src/agent/toolHander.js +async function handleOpenaiFunctionCall(url, header, body, context) { + try { + const filter_tools = ENV.USE_TOOLS.filter((i) => Object.keys(ENV.TOOLS).includes(i)).map((t) => ENV.TOOLS[t]); + if (filter_tools.length > 0) { + let tools = filter_tools.map((tool) => { + return { + "type": "function", + "function": tool.schema + }; + }); + let prompt = ENV.PROMPT["tools_prompt"]; + let call_url = url; + if (context.USER_CONFIG.FUNCTION_CALL_BASE) { + call_url = context.USER_CONFIG.FUNCTION_CALL_BASE + "/chat/completions"; + } + const call_key = context.USER_CONFIG.FUNCTION_CALL_API_KEY; + const call_headers = { ...header, ...call_key && { Authorization: `Bearer ${call_key}` } || {} }; + const options = { + fullContentExtractor: (d) => { + return d.choices?.[0]?.message; + } + }; + const call_body = { + model: context.USER_CONFIG.FUNCTION_CALL_MODEL, + tools, + tool_choice: "auto", + messages: body.messages, + stream: false + }; + if (body.messages[0].role === context.USER_CONFIG.SYSTEM_INIT_MESSAGE_ROLE) { + body.messages[0].content = prompt; + } else + body.messages.unshift({ role: "system", content: prompt }); + const call_messages = body.messages; + let call_times = ENV.FUNC_LOOP_TIMES; + const opt = {}; + const exposure_vars = ["JINA_API_KEY"]; + exposure_vars.forEach((i) => opt[i] = context.USER_CONFIG[i]); + const original_question = body.messages.at(-1).content; + let final_prompt = context.USER_CONFIG.SYSTEM_INIT_MESSAGE; + while (call_times > 0) { + const start_time = /* @__PURE__ */ new Date(); + const llm_resp = await requestChatCompletions(call_url, call_headers, call_body, context, null, null, options); + context._info.setCallInfo("call_time: " + ((/* @__PURE__ */ new Date() - start_time) / 1e3).toFixed(1) + "s"); + sendMessageToTelegramWithContext(context)("..."); + llm_resp.tool_calls = llm_resp?.tool_calls?.filter((i) => Object.keys(ENV.TOOLS).includes(i.function.name)) || []; + if (llm_resp.content?.startsWith("```json\n")) { + llm_resp.content = llm_resp.content?.match(/\{[\s\S]+\}/)[0]; + } + if (llm_resp.tool_calls.length === 0 || llm_resp.content?.startsWith?.("NO_CALL_NEEDED")) { + console.log("No need call function."); + body.messages[0].content = final_prompt; + return { type: "continue", message: "No need call function." }; + } + if (llm_resp?.content?.startsWith?.("NEED_MORE_INFO:")) { + return { type: "stop", message: llm_resp.content.substring("NEED_MORE_INFO:".length) }; + } + const funcPromise = []; + for (const func of llm_resp.tool_calls) { + const name = func.function.name; + const args = JSON.parse(func.function.arguments); + context._info.setCallInfo(`${name}: +${Object.values(args).join().substring(0, 30)}...`); + console.log("start use function: ", name); + funcPromise.push(ENV.TOOLS[name].func(args, opt)); + } + const func_resp = await Promise.all(funcPromise); + const func_time = []; + const content_text = func_resp.map((r) => { + func_time.push(r.time || ""); + return r.content || ""; + }).join("\n\n").trim(); + if (func_time.join(" ").trim()) + context._info.setCallInfo(`time: ${func_time.join()}`); + if (content_text === "") { + return { type: "continue", message: "None response in func call." }; + } + call_messages.pop(); + const tool_type = ENV.TOOLS[llm_resp.tool_calls[0].function.name].type; + const render = tools_default[tool_type].render; + call_messages.push({ + role: "user", + content: render?.(original_question, content_text) || original_question + "\n\n" + content_text + }); + if (tools_default[tool_type].prompt) + final_prompt = tools_default[tool_type].prompt; + call_times--; + } + body.messages[0].content = final_prompt; + } + return { type: "continue" }; + } catch (e) { + return { type: "error", message: e.message }; + } +} + // src/agent/openai.js function openAIKeyFromContext(context) { const length = context.USER_CONFIG.OPENAI_API_KEY.length; @@ -1388,11 +1506,10 @@ async function requestCompletionsFromOpenAI(message, prompt, history, context, o const url = `${PROXY_URL}/chat/completions`; const model = context._info?.lastStepHasFile ? context.USER_CONFIG.OPENAI_VISION_MODEL : context.USER_CONFIG.OPENAI_CHAT_MODEL; const extra_params = context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS; - const messages = [...history || []]; + const messages = [...history || [], { role: "user", content: message }]; if (prompt) { - messages.push({ role: context.USER_CONFIG.SYSTEM_INIT_MESSAGE_ROLE, content: prompt }); + messages.unshift({ role: context.USER_CONFIG.SYSTEM_INIT_MESSAGE_ROLE, content: prompt }); } - messages.push({ role: "user", content: message }); if (context._info?.lastStepHasFile) { messages.at(-1).content = [ { @@ -1415,20 +1532,21 @@ async function requestCompletionsFromOpenAI(message, prompt, history, context, o stream: onStream != null, ...!!onStream && ENV.ENABLE_SHOWTOKENINFO && { stream_options: { include_usage: true } } }; - if (prompt.includes("json") || prompt.includes("JSON")) { - body.response_format = { - "type": "json_object" - }; - } const header = { "Content-Type": "application/json", "Authorization": `Bearer ${API_KEY}` }; const options = {}; - if (extra_params.tools && extra_params.tools.length > 0) { - options.fullContentExtractor = (d) => { - return d.choices?.[0]?.message; - }; + if (message && !context._info?.lastStepHasFile && ENV.TOOLS && ENV.USE_TOOLS?.length > 0) { + const result = await handleOpenaiFunctionCall(url, header, body, context); + if (result.type === "stop") { + return result.message; + } else if (result.type === "error") { + throw new Error(result.message); + } + const resp_obj = { q: body.messages.at(-1).content }; + resp_obj.a = await requestChatCompletions(url, header, body, context, onStream, null, options); + return resp_obj; } return requestChatCompletions(url, header, body, context, onStream, null, options); } @@ -2036,7 +2154,7 @@ async function extractMessageType(message, botToken) { } let file_id = null; if (fileType == "photo") { - file_id = msg[fileType]?.sort().at(-1)?.file_id; + file_id = msg[fileType]?.at(1)?.file_id; } else { file_id = msg[fileType]?.file_id || null; } @@ -2159,6 +2277,9 @@ Token: ${Object.values(this.token_info[this.step_index]).join(" | ")}`; this.processes = this._bp_config.MODES[value][this.msg_type]; } } + updateStartTime() { + this.process_start_time[this.step_index] = Date.now(); + } initProcess(USER_CONFIG) { console.log(`Init step ${this.step_index + 1}.`); this.process_start_time.push(/* @__PURE__ */ new Date()); @@ -2276,11 +2397,15 @@ async function requestCompletionsFromLLM(text, prompt, context, llm, modifier, o history = modifierData.history; text = modifierData.text; } - const answer = await llm(text, prompt, history, context, onStream); + let answer = await llm(text, prompt, history, context, onStream); if (context._info.lastStepHasFile) { text = "[A FILE] " + text; } - if (!historyDisable && answer && typeof answer === "string") { + if (typeof answer === "object") { + text = answer.q; + answer = answer.a; + } + if (!historyDisable && answer) { history.push({ role: "user", content: text || "" }); history.push({ role: "assistant", content: answer }); await DATABASE.put(historyKey, JSON.stringify(history)).catch(console.error); @@ -2338,17 +2463,13 @@ async function chatWithLLM(text, context, modifier, pointerLLM = loadChatLLM) { } const prompt = context.USER_CONFIG.SYSTEM_INIT_MESSAGE; console.log(`[START] Chat via ${llm.name}`); - if (text && ENV.TOOLS && ENV.USE_TOOLS?.length > 0) { - const result = await handleOpenaiFunctionCall(text, context); - if (result && result instanceof Response) { - return result; - } - text = result; - } const answer = await requestCompletionsFromLLM(text, prompt, context, llm, modifier, onStream); if (!answer) { return sendMessageToTelegramWithContext(context)("None response"); } + if (answer instanceof Response) { + return answer; + } context.CURRENT_CHAT_CONTEXT.parse_mode = parseMode; if (ENV.SHOW_REPLY_BUTTON && context.CURRENT_CHAT_CONTEXT.message_id) { try { @@ -2385,65 +2506,6 @@ async function chatWithLLM(text, context, modifier, pointerLLM = loadChatLLM) { return sendMessageToTelegramWithContext(context)(errMsg); } } -async function handleOpenaiFunctionCall(text, context) { - try { - const filter_tools = ENV.USE_TOOLS.filter((i) => Object.keys(ENV.TOOLS).includes(i)).map((t) => ENV.TOOLS[t]); - if (filter_tools.length > 0) { - let tools = filter_tools.map((tool) => { - return { - "type": "function", - "function": tool.schema - }; - }); - let prompt = ENV.PROMPT["tools_prompt"]; - const bp_config = { ...context.USER_CONFIG }; - const bp_extra_params = { ...context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS }; - context.USER_CONFIG.OPENAI_CHAT_MODEL = context.USER_CONFIG.FUNCTION_CALL_MODEL || "gpt-4o"; - if (context.USER_CONFIG.FUNCTION_CALL_BASE && context.USER_CONFIG.FUNCTION_CALL_API_KEY) { - context.USER_CONFIG.OPENAI_API_BASE = context.USER_CONFIG.FUNCTION_CALL_BASE; - context.USER_CONFIG.OPENAI_API_KEY = [context.USER_CONFIG.FUNCTION_CALL_API_KEY]; - } - context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS.tools = tools; - context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS.tool_choice = "auto"; - const llm = requestCompletionsFromOpenAI; - const first_step_result = await requestCompletionsFromLLM(text, prompt, context, llm, null, null); - if (first_step_result?.content?.startsWith?.("NEED_MORE_INFO:")) { - return sendMessageToTelegramWithContext(context)(first_step_result.content.substring("NEED_MORE_INFO:".length)); - } - first_step_result.tool_calls = first_step_result?.tool_calls?.filter( - (i) => Object.keys(ENV.TOOLS).includes(i.function.name) - ); - if (!first_step_result.tool_calls || first_step_result.tool_calls?.length === 0 || first_step_result.content?.startsWith?.("NO_CALL_NEEDED")) { - console.log("No need call function."); - } else { - const options = {}; - const exposure_vars = ["JINA_API_KEY"]; - exposure_vars.forEach((i) => options[i] = context.USER_CONFIG[i]); - const func = first_step_result.tool_calls[0].function; - const function_name = func.name; - const function_args = JSON.parse(func.arguments); - console.log("start use function: ", function_name); - const last_func_result = await ENV.TOOLS[function_name].func(function_args, options); - context._info.setCallInfo(`${function_name} ${last_func_result?.time || ""}: -${Object.values(function_args)}`); - if (!last_func_result?.content?.trim()) { - return sendMessageToTelegramWithContext(context)(`None response of ${function_name}`); - } - text = ENV.TOOLS[function_name].settings?.after_render?.(text, last_func_result.content) || text + "\n" + last_func_result.content; - context.USER_CONFIG.SYSTEM_INIT_MESSAGE = ENV.TOOLS[function_name].settings?.after_prompt || bp_config.SYSTEM_INIT_MESSAGE; - } - context.USER_CONFIG.OPENAI_CHAT_MODEL = bp_config.OPENAI_CHAT_MODEL; - context.USER_CONFIG.OPENAI_API_BASE = bp_config.OPENAI_API_BASE; - context.USER_CONFIG.OPENAI_API_KEY = bp_config.OPENAI_API_KEY; - context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS = bp_extra_params; - delete context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS.tools; - delete context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS.tool_choice; - } - return text; - } catch (e) { - return sendMessageToTelegramWithContext(context)(e.message); - } -} async function chatViaFileWithLLM(context) { try { if (!context.CURRENT_CHAT_CONTEXT.message_id) { @@ -3586,7 +3648,6 @@ var jina_reader = { "required": ["url"] } }, - // before_prompt: "Only use the jina_reader function when both of these conditions are met:\n 1. The user explicitly provides a URL.\n 2. The user requests analysis or information about the content of that URL.\n\n When to use:\n • If a user asks, for \"帮我分析一下 https://example.com/news/article123\"\n • If a user says, \"I need information from this webpage: https://example.com/research/paper456\"\n • When the user's query clearly indicates they want you to read and analyze a specific online resource.\n Please remember to directly send a plain text JSON object string with only the key 'url'. For example: {'url': 'https://example.com/article'}\n\n When not to use:\n • If the user asks a general question without providing a URL.\n • If the user provides a URL but doesn't ask for analysis or information from it.\n • When you can answer the user's query using your existing knowledge.\n Then respond with 'NO_ANALYSE_NEEDED'", func: async ({ url }, { JINA_API_KEY }) => { if (!url) { throw new Error("\u53C2\u6570\u9519\u8BEF"); @@ -3595,6 +3656,7 @@ var jina_reader = { throw new Error("JINA\\_API\\_KEY \u4E0D\u5B58\u5728"); } console.log("jina-reader:", url); + const startTime = Date.now(); const result = await fetch("https://r.jina.ai/" + url, { headers: { "X-Return-Format": "text", @@ -3605,11 +3667,10 @@ var jina_reader = { if (!result.ok) { throw new Error("Error: " + (await result.json()).message); } - return { content: await result.text() }; + const time = ((Date.now() - startTime) / 1e3).toFixed(1) + "s"; + return { content: await result.text(), time }; }, - settings: { - after_prompt: '\u4F5C\u4E3A\u4E00\u4E2A\u9AD8\u6548\u7684\u5185\u5BB9\u5206\u6790\u548C\u603B\u7ED3\u52A9\u624B\uFF0C\u4F60\u7684\u4EFB\u52A1\u662F\u5BF9\u7528\u6237\u63D0\u4F9B\u7684\u7F51\u9875\u6216PDF\u5185\u5BB9\u8FDB\u884C\u5168\u9762\u800C\u7B80\u6D01\u7684\u603B\u7ED3\u3002\u8BF7\u9075\u5FAA\u4EE5\u4E0B\u6307\u5357\uFF1A\n 1. \u4ED4\u7EC6\u9605\u8BFB\u7528\u6237\u63D0\u4F9B\u7684\u5168\u90E8\u5185\u5BB9\uFF0C\u786E\u4FDD\u7406\u89E3\u4E3B\u8981\u89C2\u70B9\u548C\u5173\u952E\u4FE1\u606F\u3002\n 2. \u8BC6\u522B\u5E76\u63D0\u70BC\u51FA\u5185\u5BB9\u7684\u6838\u5FC3\u4E3B\u9898\u548C\u4E3B\u8981\u8BBA\u70B9\u3002\n 3. \u603B\u7ED3\u65F6\u5E94\u5305\u62EC\u4EE5\u4E0B\u8981\u7D20\uFF1A\n \u2022 \u5185\u5BB9\u7684\u4E3B\u8981\u76EE\u7684\u6216\u4E3B\u9898\n \u2022 \u5173\u952E\u89C2\u70B9\u6216\u8BBA\u636E\n \u2022 \u91CD\u8981\u7684\u6570\u636E\u6216\u7EDF\u8BA1\u4FE1\u606F\uFF08\u5982\u679C\u6709\uFF09\n \u2022 \u4F5C\u8005\u7684\u7ED3\u8BBA\u6216\u5EFA\u8BAE\uFF08\u5982\u679C\u9002\u7528\uFF09\n 4. \u4FDD\u6301\u5BA2\u89C2\u6027\uFF0C\u51C6\u786E\u53CD\u6620\u539F\u6587\u7684\u89C2\u70B9\uFF0C\u4E0D\u6DFB\u52A0\u4E2A\u4EBA\u89E3\u91CA\u6216\u8BC4\u8BBA\u3002\n 5. \u4F7F\u7528\u6E05\u6670\u3001\u7B80\u6D01\u7684\u8BED\u8A00\uFF0C\u907F\u514D\u4F7F\u7528\u8FC7\u4E8E\u4E13\u4E1A\u6216\u6666\u6DA9\u7684\u672F\u8BED\u3002\n 6. \u603B\u7ED3\u7684\u957F\u5EA6\u5E94\u8BE5\u662F\u539F\u6587\u768410-15%\uFF0C\u9664\u975E\u7528\u6237\u7279\u522B\u6307\u5B9A\u5176\u4ED6\u957F\u5EA6\u8981\u6C42\u3002\n 7. \u5982\u679C\u5185\u5BB9\u5305\u542B\u591A\u4E2A\u90E8\u5206\u6216\u7AE0\u8282\uFF0C\u53EF\u4EE5\u4F7F\u7528\u7B80\u77ED\u7684\u5C0F\u6807\u9898\u6765\u7EC4\u7EC7\u4F60\u7684\u603B\u7ED3\u3002\n 8. \u5982\u679C\u539F\u6587\u5305\u542B\u56FE\u8868\u6216\u56FE\u50CF\u7684\u91CD\u8981\u4FE1\u606F\uFF0C\u8BF7\u5728\u603B\u7ED3\u4E2D\u63D0\u53CA\u8FD9\u4E00\u70B9\u3002\n 9. \u5982\u679C\u5185\u5BB9\u6D89\u53CA\u65F6\u95F4\u654F\u611F\u7684\u4FE1\u606F\uFF0C\u8BF7\u5728\u603B\u7ED3\u4E2D\u6CE8\u660E\u5185\u5BB9\u7684\u53D1\u5E03\u65E5\u671F\u6216\u7248\u672C\u3002\n 10. \u5982\u679C\u539F\u6587\u5B58\u5728\u660E\u663E\u7684\u504F\u89C1\u6216\u4E89\u8BAE\u6027\u89C2\u70B9\uFF0C\u8BF7\u5728\u603B\u7ED3\u4E2D\u5BA2\u89C2\u5730\u6307\u51FA\u8FD9\u4E00\u70B9\u3002\n 11. \u603B\u7ED3\u5B8C\u6210\u540E\uFF0C\u63D0\u4F9B1-3\u4E2A\u5173\u952E\u8BCD\u6216\u77ED\u8BED\uFF0C\u6982\u62EC\u5185\u5BB9\u7684\u6838\u5FC3\u4E3B\u9898\u3002\n 12. \u5982\u679C\u7528\u6237\u8981\u6C42\uFF0C\u53EF\u4EE5\u5728\u603B\u7ED3\u7684\u6700\u540E\u6DFB\u52A0\u4E00\u4E2A\u7B80\u77ED\u7684"\u8FDB\u4E00\u6B65\u9605\u8BFB\u5EFA\u8BAE"\u90E8\u5206\u3002\n \u8BF7\u8BB0\u4F4F\uFF0C\u4F60\u7684\u76EE\u6807\u662F\u63D0\u4F9B\u4E00\u4E2A\u5168\u9762\u3001\u51C6\u786E\u3001\u6613\u4E8E\u7406\u89E3\u7684\u603B\u7ED3\uFF0C\u5E2E\u52A9\u7528\u6237\u5FEB\u901F\u628A\u63E1\u5185\u5BB9\u7684\u7CBE\u9AD3\u3002\u5982\u679C\u5185\u5BB9\u7279\u522B\u957F\u6216\u590D\u6742\uFF0C\u4F60\u53EF\u4EE5\u8BE2\u95EE\u7528\u6237\u662F\u5426\u9700\u8981\u66F4\u8BE6\u7EC6\u7684\u603B\u7ED3\u6216\u7279\u5B9A\u90E8\u5206\u7684\u6DF1\u5165\u5206\u6790\u3002' - } + type: "web_crawler" }; // main.js diff --git a/dist/timestamp b/dist/timestamp index d1767604..2fa1cdb0 100644 --- a/dist/timestamp +++ b/dist/timestamp @@ -1 +1 @@ -1722687419 +1722787231 diff --git a/src/agent/llm.js b/src/agent/llm.js index 92848e4f..aecd6d18 100644 --- a/src/agent/llm.js +++ b/src/agent/llm.js @@ -7,7 +7,6 @@ import { import {DATABASE, ENV} from '../config/env.js'; import { loadAudioLLM, loadChatLLM } from "./agents.js"; import { handleFile } from "../config/middle.js"; -import { requestCompletionsFromOpenAI } from "../agent/openai.js"; /** * @return {(function(string): number)} @@ -112,11 +111,16 @@ async function requestCompletionsFromLLM(text, prompt, context, llm, modifier, o history = modifierData.history; text = modifierData.text; } - const answer = await llm(text, prompt, history, context, onStream); + let answer = await llm(text, prompt, history, context, onStream); if (context._info.lastStepHasFile) { text = '[A FILE] ' + text; } - if (!historyDisable && answer && typeof answer === 'string') { + if (typeof answer === 'object') { + text = answer.q; + answer = answer.a; + } + + if (!historyDisable && answer) { history.push({ role: 'user', content: text || '' }); history.push({ role: 'assistant', content: answer }); await DATABASE.put(historyKey, JSON.stringify(history)).catch(console.error); @@ -193,18 +197,13 @@ export async function chatWithLLM(text, context, modifier, pointerLLM = loadChat const prompt = context.USER_CONFIG.SYSTEM_INIT_MESSAGE; console.log(`[START] Chat via ${llm.name}`); - if (text && ENV.TOOLS && ENV.USE_TOOLS?.length > 0) { - const result = await handleOpenaiFunctionCall(text, context); - if (result && result instanceof Response) { - return result; - } - text = result; - } - const answer = await requestCompletionsFromLLM(text, prompt, context, llm, modifier, onStream); if (!answer) { return sendMessageToTelegramWithContext(context)('None response'); } + if (answer instanceof Response) { + return answer; + } context.CURRENT_CHAT_CONTEXT.parse_mode = parseMode; if (ENV.SHOW_REPLY_BUTTON && context.CURRENT_CHAT_CONTEXT.message_id) { try { @@ -246,126 +245,6 @@ export async function chatWithLLM(text, context, modifier, pointerLLM = loadChat } -/** - * 处理tool - * - * @param {TelegramMessage} message - * @param {Context} context - * @return {Promise} - */ -async function handleOpenaiFunctionCall(text, context) { - try { - const filter_tools = ENV.USE_TOOLS.filter((i) => Object.keys(ENV.TOOLS).includes(i)).map((t) => ENV.TOOLS[t]); - if (filter_tools.length > 0) { - let tools = filter_tools.map((tool) => { - return { - 'type': 'function', - 'function': tool.schema, - }; - }); - - //默认使用的提示词与前缀 - let prompt = ENV.PROMPT['tools_prompt']; - // 备份现有模型与额外配置 - const bp_config = { ...context.USER_CONFIG }; - // const bp_prompt = context.USER_CONFIG.SYSTEM_INIT_MESSAGE; - const bp_extra_params = { ...context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS }; - - context.USER_CONFIG.OPENAI_CHAT_MODEL = context.USER_CONFIG.FUNCTION_CALL_MODEL || 'gpt-4o'; - if (context.USER_CONFIG.FUNCTION_CALL_BASE && context.USER_CONFIG.FUNCTION_CALL_API_KEY) { - context.USER_CONFIG.OPENAI_API_BASE = context.USER_CONFIG.FUNCTION_CALL_BASE; - context.USER_CONFIG.OPENAI_API_KEY = [context.USER_CONFIG.FUNCTION_CALL_API_KEY]; - } - context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS.tools = tools; - context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS.tool_choice = 'auto'; - - const llm = requestCompletionsFromOpenAI; - const first_step_result = await requestCompletionsFromLLM(text, prompt, context, llm, null, null); - if (first_step_result?.content?.startsWith?.('NEED_MORE_INFO:')) { - return sendMessageToTelegramWithContext(context)(first_step_result.content.substring('NEED_MORE_INFO:'.length)); - } - - // 假定function为串行，且上一步的输出是下一步的输入 - first_step_result.tool_calls = first_step_result?.tool_calls?.filter((i) => - Object.keys(ENV.TOOLS).includes(i.function.name), - ); - if ( - !first_step_result.tool_calls || - first_step_result.tool_calls?.length === 0 || - first_step_result.content?.startsWith?.('NO_CALL_NEEDED') - ) { - console.log('No need call function.'); - // return sendMessageToTelegramWithContext(context)(`Cant found function: ${function_name}`); - } else { - // let last_llm_result = first_step_result; - // let last_func_result = null; - const options = {}; - const exposure_vars = ['JINA_API_KEY']; - exposure_vars.forEach((i) => (options[i] = context.USER_CONFIG[i])); - // let function_name = null; // last_llm_result.tool_calls[0].function.name; - // let function_args = null; // JSON.parse(last_llm_result.tool_calls[0].function.arguments); - // for (const [i, func] of first_step_result.tool_calls.entries()) { - - // if (i > 0) { - // prompt = ENV.TOOLS[func.function.name].settings.before_prompt; - // message = ENV.TOOLS[func.function.name].settings.before_render(text); - // // 只保留当前需要执行的function - // context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS.tools = [ - // { - // 'type': 'function', - // 'function': ENV.TOOLS[func.function.name].schema, - // }, - // ]; - // last_llm_result = await requestCompletionsFromOpenAI(message, prompt, null, ENV, null, true); - // } - // function_name = func.function.name; - // function_args = JSON.parse(last_llm_result.tool_calls[0].function.arguments); - // context._info.setCallInfo(last_llm_result.tool_calls[0].function.arguments); - // console.log("start use function: ", function_name); - // last_func_result = await ENV.TOOLS[function_name].func(function_args, options); - // if (last_func_result instanceof Response) { - // return last_func_result; - // } - // if (!last_func_result.content) { - // return sendMessageToTelegramWithContext(context)(`None response of ${function_name}`); - // } - // // console.log(last_func_result.content); - // text = last_func_result.content; - // } - - const func = first_step_result.tool_calls[0].function; - const function_name = func.name; - const function_args = JSON.parse(func.arguments); - console.log('start use function: ', function_name); - const last_func_result = await ENV.TOOLS[function_name].func(function_args, options); - context._info.setCallInfo(`${function_name} ` + `${last_func_result?.time || ''}` + `:\n${Object.values(function_args)}`); - - if (!last_func_result?.content?.trim()) { - return sendMessageToTelegramWithContext(context)(`None response of ${function_name}`); - } - - // console.log(last_func_result.content); - text = - ENV.TOOLS[function_name].settings?.after_render?.(text, last_func_result.content) || - text + '\n' + last_func_result.content; - context.USER_CONFIG.SYSTEM_INIT_MESSAGE = - ENV.TOOLS[function_name].settings?.after_prompt || bp_config.SYSTEM_INIT_MESSAGE; - } - - context.USER_CONFIG.OPENAI_CHAT_MODEL = bp_config.OPENAI_CHAT_MODEL; - context.USER_CONFIG.OPENAI_API_BASE = bp_config.OPENAI_API_BASE; - context.USER_CONFIG.OPENAI_API_KEY = bp_config.OPENAI_API_KEY; - context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS = bp_extra_params; - // - delete context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS.tools; - delete context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS.tool_choice; - } - return text; - } catch (e) { - return sendMessageToTelegramWithContext(context)(e.message); - } -} - export async function chatViaFileWithLLM(context) { try { if (!context.CURRENT_CHAT_CONTEXT.message_id) { diff --git a/src/agent/openai.js b/src/agent/openai.js index 0b74b46a..1a50dcc6 100644 --- a/src/agent/openai.js +++ b/src/agent/openai.js @@ -1,7 +1,7 @@ import "../types/context.js"; import {requestChatCompletions} from "./request.js"; import {ENV} from "../config/env.js"; - +import { handleOpenaiFunctionCall } from "../agent/toolHander.js"; /** * @param {ContextType} context @@ -40,12 +40,11 @@ export async function requestCompletionsFromOpenAI(message, prompt, history, con ? context.USER_CONFIG.OPENAI_VISION_MODEL : context.USER_CONFIG.OPENAI_CHAT_MODEL; const extra_params = context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS; - const messages = [...(history || [])]; + const messages = [...(history || []), { role: 'user', content: message }]; if (prompt) { - messages.push({ role: context.USER_CONFIG.SYSTEM_INIT_MESSAGE_ROLE, content: prompt }); + messages.unshift({ role: context.USER_CONFIG.SYSTEM_INIT_MESSAGE_ROLE, content: prompt }); } - messages.push({ role: 'user', content: message }); // 优先取原始文件兼容claude if (context._info?.lastStepHasFile) { @@ -70,11 +69,6 @@ export async function requestCompletionsFromOpenAI(message, prompt, history, con stream: onStream != null, ...(!!onStream && ENV.ENABLE_SHOWTOKENINFO && { stream_options: { include_usage: true } }), }; - if (prompt.includes('json') || prompt.includes('JSON')) { - body.response_format = { - 'type': 'json_object', - }; - } const header = { 'Content-Type': 'application/json', @@ -82,10 +76,17 @@ export async function requestCompletionsFromOpenAI(message, prompt, history, con }; const options = {}; - if (extra_params.tools && extra_params.tools.length > 0) { - options.fullContentExtractor = (d) => { - return d.choices?.[0]?.message; - }; + if (message && !context._info?.lastStepHasFile && ENV.TOOLS && ENV.USE_TOOLS?.length > 0) { + const result = await handleOpenaiFunctionCall(url, header, body, context); + if (result.type === 'stop') { + return result.message; + } else if (result.type === 'error') { + throw new Error(result.message); + } + const resp_obj = { q: body.messages.at(-1).content }; // 修正问题内容 + resp_obj.a = await requestChatCompletions(url, header, body, context, onStream, null, options); + return resp_obj; + } return requestChatCompletions(url, header, body, context, onStream, null, options); diff --git a/src/agent/request.js b/src/agent/request.js index 1f1ae90e..7584bf9f 100644 --- a/src/agent/request.js +++ b/src/agent/request.js @@ -1,7 +1,6 @@ import '../types/context.js'; import { ENV } from '../config/env.js'; import { Stream } from './stream.js'; -import { fetchWithRetry } from '../utils/utils.js'; /** * @@ -107,8 +106,10 @@ export async function requestChatCompletions(url, header, body, context, onStrea if (ENV.DEBUG_MODE) { console.log(`url:\n${url}\nheader:\n${JSON.stringify(header)}\nbody:\n${JSON.stringify(body, null, 2)}`); } + // 排除 function call耗时 + context._info.updateStartTime(); - const resp = await fetchWithRetry(url, { + const resp = await fetch(url, { method: 'POST', headers: header, body: JSON.stringify(body), @@ -126,7 +127,7 @@ export async function requestChatCompletions(url, header, body, context, onStrea const stream = options.streamBuilder(resp, controller); let contentFull = ''; let lengthDelta = 0; - let updateStep = 10; + let updateStep = 20; let msgPromise = null; let lastChunk = null; let usage = null; diff --git a/src/agent/toolHander.js b/src/agent/toolHander.js new file mode 100644 index 00000000..d0db33e9 --- /dev/null +++ b/src/agent/toolHander.js @@ -0,0 +1,127 @@ +import { requestChatCompletions } from './request.js'; +import { ENV } from '../config/env.js'; +import { sendMessageToTelegramWithContext } from '../telegram/telegram.js'; +import tools_settings from '../prompt/tools.js'; + +/** + * 处理tool + * + * @param {TelegramMessage} message + * @param {Context} context + * @return {Promise} + */ +export async function handleOpenaiFunctionCall(url, header, body, context) { + try { + const filter_tools = ENV.USE_TOOLS.filter((i) => Object.keys(ENV.TOOLS).includes(i)).map((t) => ENV.TOOLS[t]); + if (filter_tools.length > 0) { + let tools = filter_tools.map((tool) => { + return { + 'type': 'function', + 'function': tool.schema, + }; + }); + + //默认使用的提示词与前缀 + let prompt = ENV.PROMPT['tools_prompt']; + let call_url = url; + if (context.USER_CONFIG.FUNCTION_CALL_BASE) { + call_url = context.USER_CONFIG.FUNCTION_CALL_BASE + '/chat/completions'; + } + const call_key = context.USER_CONFIG.FUNCTION_CALL_API_KEY; + const call_headers = { ...header, ...((call_key && { Authorization: `Bearer ${call_key}` }) || {}) }; + + const options = { + fullContentExtractor: (d) => { + return d.choices?.[0]?.message; + }, + }; + + const call_body = { + model: context.USER_CONFIG.FUNCTION_CALL_MODEL, + tools, + tool_choice: 'auto', + messages: body.messages, + stream: false, + }; + + // if (prompt.includes('json') || prompt.includes('JSON')) { + // body.response_format = { + // 'type': 'json_object', + // }; + // } + if (body.messages[0].role === context.USER_CONFIG.SYSTEM_INIT_MESSAGE_ROLE) { + body.messages[0].content = prompt; + } else body.messages.unshift({ role: 'system', content: prompt }); + + const call_messages = body.messages; + let call_times = ENV.FUNC_LOOP_TIMES; //最多循环调用次数 + const opt = {}; + const exposure_vars = ['JINA_API_KEY']; + exposure_vars.forEach((i) => (opt[i] = context.USER_CONFIG[i])); + const original_question = body.messages.at(-1).content; + let final_prompt = context.USER_CONFIG.SYSTEM_INIT_MESSAGE; + while (call_times > 0) { + const start_time = new Date(); + const llm_resp = await requestChatCompletions(call_url, call_headers, call_body, context, null, null, options); + context._info.setCallInfo('call_time: ' + ((new Date() - start_time) / 1000).toFixed(1) + 's'); + sendMessageToTelegramWithContext(context)('...'); + llm_resp.tool_calls = + llm_resp?.tool_calls?.filter((i) => Object.keys(ENV.TOOLS).includes(i.function.name)) || []; + if (llm_resp.content?.startsWith('```json\n')) { + llm_resp.content = llm_resp.content?.match(/\{[\s\S]+\}/)[0]; + } + + if (llm_resp.tool_calls.length === 0 || llm_resp.content?.startsWith?.('NO_CALL_NEEDED')) { + console.log('No need call function.'); + body.messages[0].content = final_prompt; + return { type: 'continue', message: 'No need call function.' }; + // return false; + } + + if (llm_resp?.content?.startsWith?.('NEED_MORE_INFO:')) { + return { type: 'stop', message: llm_resp.content.substring('NEED_MORE_INFO:'.length) }; + // return llm_resp.content.substring('NEED_MORE_INFO:'.length); + } + + const funcPromise = []; + for (const func of llm_resp.tool_calls) { + const name = func.function.name; + const args = JSON.parse(func.function.arguments); + context._info.setCallInfo(`${name}:\n${Object.values(args).join().substring(0,30)}...`); + console.log('start use function: ', name); + funcPromise.push(ENV.TOOLS[name].func(args, opt)); + } + + const func_resp = await Promise.all(funcPromise); + const func_time = []; + const content_text = func_resp + .map((r) => { + func_time.push(r.time || ''); + return r.content || ''; + }) + .join('\n\n') + .trim(); + if(func_time.join(' ').trim()) context._info.setCallInfo(`time: ${func_time.join()}`); + if(content_text === '') { + return { type: 'continue', message: 'None response in func call.' }; + // throw new Error(llm_resp.content.substring('None response in func call.')); + } + call_messages.pop(); + const tool_type = ENV.TOOLS[llm_resp.tool_calls[0].function.name].type; + const render = tools_settings[tool_type].render; + call_messages.push({ + role: 'user', + content: render?.(original_question, content_text) || original_question + '\n\n' + content_text, + }); + if (tools_settings[tool_type].prompt) final_prompt = tools_settings[tool_type].prompt; + call_times--; + } + body.messages[0].content = final_prompt; + } + return { type: 'continue' }; + // return false; + } catch (e) { + // throw new Error(e.message); + return { type: 'error', message: e.message }; + } +} diff --git a/src/config/env.js b/src/config/env.js index ab0749d1..78358e79 100644 --- a/src/config/env.js +++ b/src/config/env.js @@ -142,7 +142,7 @@ export class UserConfig { CURRENT_MODE = 'default'; JINA_API_KEY = ''; // openai格式调用FUNCTION CALL参数 - FUNCTION_CALL_MODEL = ''; + FUNCTION_CALL_MODEL = 'gpt-4o-mini'; FUNCTION_CALL_API_KEY = ''; FUNCTION_CALL_BASE = ''; } @@ -276,6 +276,7 @@ class Environment { // 需要使用的函数 当前仅 duckduckgo_search 和jina_reader // '["duckduckgo_search", "jina_reader"]' USE_TOOLS = []; + FUNC_LOOP_TIMES = 1; } export const ENV_KEY_MAPPER = { diff --git a/src/config/middle.js b/src/config/middle.js index 2c233bed..978f9190 100644 --- a/src/config/middle.js +++ b/src/config/middle.js @@ -50,8 +50,8 @@ async function extractMessageType(message, botToken) { } let file_id = null; if (fileType == 'photo') { - // 取最后一个文件，像素最高 - file_id = msg[fileType]?.sort().at(-1)?.file_id /*|| msg[fileType]?.file_id*/; + // 取第二个 + file_id = msg[fileType]?.at(1)?.file_id /*|| msg[fileType]?.file_id*/; } else { file_id = msg[fileType]?.file_id || null; } @@ -191,6 +191,9 @@ export class MiddleInfo { this.processes = this._bp_config.MODES[value][this.msg_type]; } // else this.processes[this.step_index][name] = value; } + updateStartTime() { + this.process_start_time[this.step_index] = Date.now(); + } initProcess(USER_CONFIG) { console.log(`Init step ${this.step_index + 1}.`); diff --git a/src/prompt/tools.js b/src/prompt/tools.js new file mode 100644 index 00000000..3f6eeeae --- /dev/null +++ b/src/prompt/tools.js @@ -0,0 +1,12 @@ +export default { + search: { + prompt: + '作为智能助手，请按照以下步骤有效分析并提取我提供的搜索结果，以简洁明了的方式回答我的问题：\n\n1. 阅读和评估：仔细阅读所有搜索结果，识别并优先获取来自可靠和最新来源的信息。考虑因素包括官方来源、知名机构以及信息的更新时间。\n\n2. 提取关键信息：\n • *汇率查询*：提供最新汇率并进行必要的换算。\n • *天气查询*：提供具体地点和时间的天气预报。\n • *事实性问题*：找出权威回答。\n\n3. 简洁回答：对提取的信息进行综合分析，给出简明扼要的回答。\n\n4. 识别不确定性：如果信息存在矛盾或不确定性，请解释可能原因。\n\n5. 说明信息不足：如果搜索结果无法完全回答问题，指出需要的额外信息。\n\n6. 用户友好：使用简单易懂的语言，必要时提供简短解释，确保回答易于理解。\n\n7. 附加信息：根据需要提供额外相关信息或建议，以增强回答的价值。\n\n8. 来源标注：在回答中清晰标注信息来源，包括来源网站或机构名称及数据的发布或更新时间。\n\n9. 参考列表：如果引用了多个来源，在回答最后提供简短的参考列表，列出主要信息来源。\n\n请确保目标是提供最新、最相关和最有用的信息，直接回应我的问题。避免冗长的细节，聚焦于我最关心的核心答案，并通过可靠的来源增强回答的可信度。', + render: (question, result) => `问题:\n${question}\n\n搜索结果:\n${result}`, + }, + web_crawler: { + prompt: + '作为一个高效的内容分析和总结助手，你的任务是对用户提供的网页或PDF内容进行全面而简洁的总结。请遵循以下指南：\n 1. 仔细阅读用户提供的全部内容，确保理解主要观点和关键信息。\n 2. 识别并提炼出内容的核心主题和主要论点。\n 3. 总结时应包括以下要素：\n • 内容的主要目的或主题\n • 关键观点或论据\n • 重要的数据或统计信息（如果有）\n • 作者的结论或建议（如果适用）\n 4. 保持客观性，准确反映原文的观点，不添加个人解释或评论。\n 5. 使用清晰、简洁的语言，避免使用过于专业或晦涩的术语。\n 6. 总结的长度应该是原文的10-15%，除非用户特别指定其他长度要求。\n 7. 如果内容包含多个部分或章节，可以使用简短的小标题来组织你的总结。\n 8. 如果原文包含图表或图像的重要信息，请在总结中提及这一点。\n 9. 如果内容涉及时间敏感的信息，请在总结中注明内容的发布日期或版本。\n 10. 如果原文存在明显的偏见或争议性观点，请在总结中客观地指出这一点。\n 11. 总结完成后，提供1-3个关键词或短语，概括内容的核心主题。\n 12. 如果用户要求，可以在总结的最后添加一个简短的"进一步阅读建议"部分。\n 请记住，你的目标是提供一个全面、准确、易于理解的总结，帮助用户快速把握内容的精髓。如果内容特别长或复杂，你可以询问用户是否需要更详细的总结或特定部分的深入分析。', + render: (question, result) => `${question && question + '\n'}网页内容:\n${result}`, + }, +}; diff --git a/src/telegram/message.js b/src/telegram/message.js index dec5dfdc..34cfaddb 100644 --- a/src/telegram/message.js +++ b/src/telegram/message.js @@ -398,124 +398,6 @@ async function msgChatWithLLM(message, context) { return new Response('success', { status: 200 }); } -// /** -// * 处理tool -// * -// * @param {TelegramMessage} message -// * @param {Context} context -// * @return {Promise} -// */ -// async function handleOpenaiFunctionCall(message, context) { -// try { -// let text = (message.text || message.caption || '').trim(); -// if (ENV.EXTRA_MESSAGE_CONTEXT && context.SHARE_CONTEXT?.extraMessageContext?.text) { -// text = -// context.SHARE_CONTEXT.extraMessageContext.text || -// context.SHARE_CONTEXT.extraMessageContext.caption + '\n' + text; -// } - -// if (text && ENV.TOOLS && ENV.USE_TOOLS?.length > 0) { -// const filter_tools = ENV.USE_TOOLS.filter((i) => Object.keys(ENV.TOOLS).includes(i)).map((t) => ENV.TOOLS[t]); -// if (filter_tools.length > 0) { -// let tools = filter_tools.map((tool) => { -// return { -// 'type': 'function', -// 'function': tool.schema, -// }; -// }); - -// //默认使用的提示词与前缀 -// let prompt = ENV.PROMPT["tools_prompt"]; -// let message = text; -// // 备份现有模型与额外配置 -// const bp_config = {...context.USER_CONFIG}; -// // const bp_prompt = context.USER_CONFIG.SYSTEM_INIT_MESSAGE; -// const bp_extra_params = { ...context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS }; - -// context.USER_CONFIG.OPENAI_CHAT_MODEL = context.USER_CONFIG.FUNCTION_CALL_MODEL || 'gpt-4o'; -// if (context.USER_CONFIG.FUNCTION_CALL_BASE && context.USER_CONFIG.FUNCTION_CALL_API_KEY) { -// context.USER_CONFIG.OPENAI_API_BASE = context.USER_CONFIG.FUNCTION_CALL_BASE; -// context.USER_CONFIG.OPENAI_API_KEY = [context.USER_CONFIG.FUNCTION_CALL_API_KEY]; -// } -// context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS.tools = tools; -// context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS.tool_choice = 'auto'; - -// const first_step_result = await requestCompletionsFromOpenAI(message, prompt, null, context, null, true); - -// // 假定function为串行，且上一步的输出是下一步的输入 -// first_step_result.tool_calls = first_step_result?.tool_calls?.filter((i) => -// Object.keys(ENV.TOOLS).includes(i.function.name), -// ); -// if (!first_step_result.tool_calls || first_step_result.tool_calls?.length === 0) { -// console.log('No need call function.'); -// // return sendMessageToTelegramWithContext(context)(`Cant found function: ${function_name}`); -// } else { -// // let last_llm_result = first_step_result; -// // let last_func_result = null; -// const options = {}; -// const exposure_vars = ['JINA_API_KEY']; -// exposure_vars.forEach((i) => (options[i] = context.USER_CONFIG[i])); -// // let function_name = null; // last_llm_result.tool_calls[0].function.name; -// // let function_args = null; // JSON.parse(last_llm_result.tool_calls[0].function.arguments); -// // for (const [i, func] of first_step_result.tool_calls.entries()) { - -// // if (i > 0) { -// // prompt = ENV.TOOLS[func.function.name].settings.before_prompt; -// // message = ENV.TOOLS[func.function.name].settings.before_render(text); -// // // 只保留当前需要执行的function -// // context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS.tools = [ -// // { -// // 'type': 'function', -// // 'function': ENV.TOOLS[func.function.name].schema, -// // }, -// // ]; -// // last_llm_result = await requestCompletionsFromOpenAI(message, prompt, null, ENV, null, true); -// // } -// // function_name = func.function.name; -// // function_args = JSON.parse(last_llm_result.tool_calls[0].function.arguments); -// // context._info.setCallInfo(last_llm_result.tool_calls[0].function.arguments); -// // console.log("start use function: ", function_name); -// // last_func_result = await ENV.TOOLS[function_name].func(function_args, options); -// // if (last_func_result instanceof Response) { -// // return last_func_result; -// // } -// // if (!last_func_result.content) { -// // return sendMessageToTelegramWithContext(context)(`None response of ${function_name}`); -// // } -// // // console.log(last_func_result.content); -// // text = last_func_result.content; -// // } - -// const func = first_step_result.tool_calls[0].function; -// const function_name = func.name; -// const function_args = JSON.parse(func.arguments); -// context._info.setCallInfo(`${function_name}:\n${Object.values(function_args)}`); -// console.log('start use function: ', function_name); -// const last_func_result = await ENV.TOOLS[function_name].func(function_args, options); - -// if (!last_func_result?.content?.trim()) { -// return sendMessageToTelegramWithContext(context)(`None response of ${function_name}`); -// } - -// // console.log(last_func_result.content); -// text = last_func_result.content.trim(); -// context.USER_CONFIG.SYSTEM_INIT_MESSAGE = ENV.TOOLS[function_name].settings?.after_prompt || bp_prompt; -// text = ENV.TOOLS[function_name].settings?.after_render?.(text) || text; -// } -// context.USER_CONFIG = bp_config; -// context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS = bp_extra_params; -// // -// delete context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS.tools; -// delete context.USER_CONFIG.OPENAI_API_EXTRA_PARAMS.tool_choice; -// } -// } -// message.text = text; -// return null; - -// } catch (e) { -// return sendMessageToTelegramWithContext(context)(e.message); -// } -// } /** diff --git a/src/tools/duckduckgo.js b/src/tools/duckduckgo.js index 14e80550..110a3e59 100644 --- a/src/tools/duckduckgo.js +++ b/src/tools/duckduckgo.js @@ -39,15 +39,5 @@ export const duckduckgo_search = { return { content, time }; }, - settings: { - // before_model: 'gpt-4o-mini', - // after_model: '', - // before_prompt: "你是一个智能助手，具备广泛的知识库，同时也能在需要时调用搜索引擎获取实时信息。你的主要任务是:\n\n 1. 仔细分析用户的问题，判断是否需要获取实时或最新信息，不要猜测答案，如果你不确定，请调用搜索函数。\n 2. 识别用户查询中可能需要实时数据的关键词，如\"现在\"、\"最新\"、\"实时\"、\"今天\"等，如果用户明确提出要求联网:\"搜一下, 搜搜， search\"，请调用搜索函数。\n 3. 对于以下类型的查询，通常需要获取最新信息：\n - 实时新闻和当前事件\n - 天气预报\n - 当前时间\n - 股票价格和市场数据\n - 体育比分和赛事结果\n - 热门话题和趋势\n - 最新发布的内容（如电影、音乐、游戏等）\n 4. 如果问题涉及具体日期、数字或需要即时计算，也需要调用函数进行搜索\n 5. 对于历史事实、科学知识、常识性问题，优先使用你的内置知识回答。\n 6. 如果不确定信息的时效性或准确性，宁可调用搜索函数，获取最新数据。\n 7. 当你确定需要获取实时信息时，执行以下步骤：\n a. 生成3-4个最相关的搜索关键词。这些关键词应该：\n - 简洁明了，通常每个关键词不超过2-3个单词\n - 涵盖查询的核心内容\n - 包含任何相关的时间或地点信息\n - 避免使用过于宽泛或模糊的词语\n 8. 在你的回答中，清晰地表明哪些信息是基于实时查询，哪些是来自你的知识库。\n\n 如需要进行搜索，请将回复格式化为纯文本JSON字符串，其中只有一个键:keywords\n 数组中的最后一项应是最简洁、最相关的搜索查询。\n Examples:\n 1. For \"你能做什么？\", respond with 'NO_SEARCH_NEEDED'.\n 2. For \"珠三角是否包括佛山？\", respond with:\n {\"keywords\":[\"珠三角\", \"佛山\", \"广东省\", \"珠江三角洲 包括 佛山\"]}\n 3. For \"2023年世界杯冠军是谁？\", respond with:\n {\"keywords\":[\"2023年\", \"世界杯\", \"冠军\", \"2023 世界杯 冠军队\"]}", - - // before_render: (question) => `Analyze and generate search queries if needed for: ${question}`, - - after_prompt: "作为智能助手，请按照以下步骤有效分析并提取我提供的搜索结果，以简洁明了的方式回答我的问题：\n\n1. 阅读和评估：仔细阅读所有搜索结果，识别并优先获取来自可靠和最新来源的信息。考虑因素包括官方来源、知名机构以及信息的更新时间。\n\n2. 提取关键信息：\n • *汇率查询*：提供最新汇率并进行必要的换算。\n • *天气查询*：提供具体地点和时间的天气预报。\n • *事实性问题*：找出权威回答。\n\n3. 简洁回答：对提取的信息进行综合分析，给出简明扼要的回答。\n\n4. 识别不确定性：如果信息存在矛盾或不确定性，请解释可能原因。\n\n5. 说明信息不足：如果搜索结果无法完全回答问题，指出需要的额外信息。\n\n6. 用户友好：使用简单易懂的语言，必要时提供简短解释，确保回答易于理解。\n\n7. 附加信息：根据需要提供额外相关信息或建议，以增强回答的价值。\n\n8. 来源标注：在回答中清晰标注信息来源，包括来源网站或机构名称及数据的发布或更新时间。\n\n9. 参考列表：如果引用了多个来源，在回答最后提供简短的参考列表，列出主要信息来源。\n\n请确保目标是提供最新、最相关和最有用的信息，直接回应我的问题。避免冗长的细节，聚焦于我最关心的核心答案，并通过可靠的来源增强回答的可信度。", - after_render: (question, result) => `问题：${question}\n\n搜索结果：${result}`, - }, - // after_history_length: 0, + type: 'search' }; \ No newline at end of file diff --git a/src/tools/jina.js b/src/tools/jina.js index b896ea64..37de5a84 100644 --- a/src/tools/jina.js +++ b/src/tools/jina.js @@ -16,8 +16,6 @@ export const jina_reader = { }, }, - // before_prompt: "Only use the jina_reader function when both of these conditions are met:\n 1. The user explicitly provides a URL.\n 2. The user requests analysis or information about the content of that URL.\n\n When to use:\n • If a user asks, for \"帮我分析一下 https://example.com/news/article123\"\n • If a user says, \"I need information from this webpage: https://example.com/research/paper456\"\n • When the user's query clearly indicates they want you to read and analyze a specific online resource.\n Please remember to directly send a plain text JSON object string with only the key 'url'. For example: {'url': 'https://example.com/article'}\n\n When not to use:\n • If the user asks a general question without providing a URL.\n • If the user provides a URL but doesn't ask for analysis or information from it.\n • When you can answer the user's query using your existing knowledge.\n Then respond with 'NO_ANALYSE_NEEDED'", - func: async ({ url }, { JINA_API_KEY }) => { if (!url) { throw new Error('参数错误'); @@ -26,6 +24,7 @@ export const jina_reader = { throw new Error('JINA\\_API\\_KEY 不存在'); } console.log('jina-reader:', url); + const startTime = Date.now(); const result = await fetch('https://r.jina.ai/' + url, { headers: { 'X-Return-Format': 'text', @@ -36,10 +35,9 @@ export const jina_reader = { if (!result.ok) { throw new Error('Error: ' + (await result.json()).message); } - return {content: await result.text()}; - }, - - settings: { - after_prompt: "作为一个高效的内容分析和总结助手，你的任务是对用户提供的网页或PDF内容进行全面而简洁的总结。请遵循以下指南：\n 1. 仔细阅读用户提供的全部内容，确保理解主要观点和关键信息。\n 2. 识别并提炼出内容的核心主题和主要论点。\n 3. 总结时应包括以下要素：\n • 内容的主要目的或主题\n • 关键观点或论据\n • 重要的数据或统计信息（如果有）\n • 作者的结论或建议（如果适用）\n 4. 保持客观性，准确反映原文的观点，不添加个人解释或评论。\n 5. 使用清晰、简洁的语言，避免使用过于专业或晦涩的术语。\n 6. 总结的长度应该是原文的10-15%，除非用户特别指定其他长度要求。\n 7. 如果内容包含多个部分或章节，可以使用简短的小标题来组织你的总结。\n 8. 如果原文包含图表或图像的重要信息，请在总结中提及这一点。\n 9. 如果内容涉及时间敏感的信息，请在总结中注明内容的发布日期或版本。\n 10. 如果原文存在明显的偏见或争议性观点，请在总结中客观地指出这一点。\n 11. 总结完成后，提供1-3个关键词或短语，概括内容的核心主题。\n 12. 如果用户要求，可以在总结的最后添加一个简短的\"进一步阅读建议\"部分。\n 请记住，你的目标是提供一个全面、准确、易于理解的总结，帮助用户快速把握内容的精髓。如果内容特别长或复杂，你可以询问用户是否需要更详细的总结或特定部分的深入分析。", + const time = ((Date.now() - startTime) / 1000).toFixed(1) + 's'; + return { content: await result.text(), time }; }, + + type: 'web_crawler' }; \ No newline at end of file diff --git a/src/utils/utils.js b/src/utils/utils.js index 21a48a0e..680c6456 100644 --- a/src/utils/utils.js +++ b/src/utils/utils.js @@ -114,6 +114,7 @@ function fetchWithRetryFunc() { const DEFAULT_RETRY_AFTER = 10; return async (url, options, retries = MAX_RETRIES, delayMs = RETRY_DELAY_MS) => { + let errorMsg = ""; while (retries > 0) { try { const parsedUrl = new URL(url); @@ -140,24 +141,25 @@ function fetchWithRetryFunc() { if (retries < MAX_RETRIES) console.log(`[DONE] after ${MAX_RETRIES - retries} times`); return resp; } - const clone_resp = await resp.clone().json(); - console.log(`${JSON.stringify(clone_resp)}`); + const clone_resp = await resp.clone().text(); + console.error(`Error fetch: ${clone_resp}`); if (resp.status === 429) { const retryAfter = (resp.headers.get('Retry-After')) || DEFAULT_RETRY_AFTER; // const retryAfter = resp?.parameters?.retry_after || resp.headers.get('Retry-After') || DEFAULT_RETRY_AFTER; status429RetryTime[domain] = Date.now() + 1000 * retryAfter; return resp; } else { - return resp; + throw new Error(clone_resp); } } catch (error) { + errorMsg = error.message; console.log(`Request failed, retry after ${delayMs / 1000} s: ${error}`); } await delay(delayMs); delayMs *= RETRY_MULTIPLIER; retries--; } - throw new Error('Failed after maximum retries'); + throw new Error(`Failed after maximum retries: ${errorMsg}`); }; }