@keo-ai/axiom 0.2.8 → 0.2.10

package/README.md

@@ -177,14 +177,17 @@ for await (const chunk of LLM.streamPredict({ model: 'qwen-max', prompt: '讲个
   if (chunk.type === 'reasoning') {
     process.stdout.write(chunk.delta);  // 推理过程
   }
-  if (chunk.type === 'finish' && chunk.usage) {
+  if (chunk.type === 'finish') {
+    console.log('Finish:', chunk.finishReason, chunk.sawDone);
     console.log('Token usage:', chunk.usage);
-    // { promptTokens, completionTokens, totalTokens, cachedPromptTokens? }
   }
 }
 ```
 
-> 💡 流式调用自动启用 `stream_options: { include_usage: true }`，`finish` 事件携带完整的 token 消耗统计。usage 可能与 `finish_reason` 在同一个 SSE chunk，也可能在独立的 chunk（`choices: []`）中返回，两种情况均已兼容。
+> 💡 流式调用自动启用 `stream_options: { include_usage: true }`。`finish` 事件会返回 `finishReason`、
+> `sawDone`、`sawFinishReason` 和 token 消耗统计。usage 可能与 `finish_reason` 在同一个 SSE chunk，也可能在独立的
+> chunk（`choices: []`）中返回，两种情况均已兼容。若连接结束前既没有 `[DONE]` 也没有非空
+> `finish_reason`，流会抛出异常，避免把提前 EOF 当作成功响应。
 
 ### 模型列表
 
@@ -193,12 +196,14 @@ for await (const chunk of LLM.streamPredict({ model: 'qwen-max', prompt: '讲个
 | 模型 | 推理深度 | json 模式 | 说明 |
 |------|----------|-----------|------|
 | `qwen-max` | ❌ 不支持 | ✅ | **默认模型** |
-| `qwen3.7-max` | ✅ 映射支持 | ✅ | `low`/`medium`/`high` 映射为 `enable_thinking` |
+| `qwen3.8-flash` | ✅ 映射支持 | ✅ | `low`/`medium`/`high` 映射为 `enable_thinking` |
+| `qwen3.7-max` | ✅ 映射支持 | ✅ | 同上 |
 | `qwen3.7-flash` | ✅ 映射支持 | ✅ | 同上 |
 | `qwen-plus` | ✅ 映射支持 | ✅ | 同上 |
 | `qwen-turbo` | ✅ 映射支持 | ✅ | 同上 |
 | `deepseek-v4-pro` | ✅ 映射支持 | ✅ | 同上（百炼走「混合思考模式」，由 `enable_thinking` 控制） |
 | `deepseek-v4-flash` | ✅ 映射支持 | ✅ | 同上 |
+| `deepseek-v4.1-flash` | ✅ 映射支持 | ✅ | 同上（默认开启思考） |
 | `kimi-k2.6` | ✅ 映射支持 | ✅ | `low`/`medium`/`high` 映射为 `extra_body.thinking.type` |
 | `qwq-plus` | ❌ 不支持 | ✅ | 固定推理行为，不可调节 |
 | `glm-5.1` | ❌ 不支持 | ❌ | 不支持推理参数，不支持 json 模式 |
@@ -224,6 +229,15 @@ try {
 }
 ```
 
+`responseFormat: 'json'` 时输出为空或无法 `JSON.parse`，抛出的 `Error.message` 会带上 `model`、`finish_reason`、
+原文长度和模型原文，调用方直接打日志即可排查：`finish_reason=length` 表示被 `maxTokens` 截断，`stop` 表示模型本身输出了非法
+JSON，`content_filter` 表示被内容审核拦截。
+
+```
+JSON parse failed: Unterminated string in JSON at position 1118 (model=qwen3.8-flash, finish_reason=length, raw_length=1118)
+raw: {"classification":"normal","reason":"...
+```
+
 ---
 
 ## Function Call Loop 模块

package/dist/llm_provider/bailian.js

@@ -60,7 +60,16 @@ class BailianProvider {
         // 通过 extra_body.enable_thinking 控制思考开关（low 关闭，medium/high 开启）；
         // Kimi（Moonshot）则通过 extra_body.thinking.type 实现（默认开启）。
         if (body.reasoning_effort !== undefined) {
-            if (['qwen-plus', 'qwen-turbo', 'qwen3.7-max', 'qwen3.7-flash', 'deepseek-v4-pro', 'deepseek-v4-flash'].includes(body.model)) {
+            if ([
+                'qwen-plus',
+                'qwen-turbo',
+                'qwen3.8-flash',
+                'qwen3.7-max',
+                'qwen3.7-flash',
+                'deepseek-v4-pro',
+                'deepseek-v4-flash',
+                'deepseek-v4.1-flash',
+            ].includes(body.model)) {
                 const { reasoning_effort } = body, rest = __rest(body, ["reasoning_effort"]);
                 return Object.assign(Object.assign({}, rest), { extra_body: Object.assign(Object.assign({}, ((_c = rest.extra_body) !== null && _c !== void 0 ? _c : {})), { enable_thinking: reasoning_effort !== 'low' }) });
             }
@@ -82,7 +91,7 @@ class BailianProvider {
      */
     generate(request) {
         return __awaiter(this, void 0, void 0, function* () {
-            var _a, _b;
+            var _a, _b, _c;
             const body = this.adaptRequest(Object.assign(Object.assign({}, this.buildRequestBody(request)), { stream: false }));
             const data = yield (0, index_1.callChatCompletions)(this.config.baseUrl, this.config.apiKey, body, this.name);
             const choice = data.choices[0];
@@ -98,6 +107,7 @@ class BailianProvider {
                     }
                     : undefined,
                 model: (_b = data.model) !== null && _b !== void 0 ? _b : body.model,
+                finishReason: (_c = choice.finish_reason) !== null && _c !== void 0 ? _c : undefined,
             };
         });
     }
@@ -109,7 +119,7 @@ class BailianProvider {
      */
     stream(request) {
         return __asyncGenerator(this, arguments, function* stream_1() {
-            var _a, _b, _c;
+            var _a;
             const url = `${this.config.baseUrl}/chat/completions`;
             const body = this.adaptRequest(Object.assign(Object.assign({}, this.buildRequestBody(request)), { stream: true, stream_options: { include_usage: true } }));
             let response;
@@ -140,6 +150,53 @@ class BailianProvider {
             // usage 可能与 finish_reason 在同一个 chunk，也可能在独立的 chunk（choices: []）中返回。
             // 统一在此收集，流结束后一次性 yield finish 事件。
             let streamUsage;
+            let finishReason;
+            let sawDone = false;
+            let sawFinishReason = false;
+            function* processSseLine(line) {
+                var _a, _b;
+                const trimmed = line.trim();
+                if (!trimmed || !trimmed.startsWith('data:'))
+                    return;
+                const data = trimmed.slice(5).trimStart();
+                if (!data)
+                    return;
+                if (data === '[DONE]') {
+                    sawDone = true;
+                    return;
+                }
+                let parsed;
+                try {
+                    parsed = JSON.parse(data);
+                }
+                catch (cause) {
+                    const message = cause instanceof Error ? cause.message : String(cause);
+                    throw new Error(`[bailian] Malformed SSE JSON: ${message}`);
+                }
+                // 从任意 chunk 中收集 usage（包括 choices 为空的独立 usage chunk）
+                if (parsed.usage) {
+                    streamUsage = {
+                        promptTokens: parsed.usage.prompt_tokens,
+                        completionTokens: parsed.usage.completion_tokens,
+                        totalTokens: parsed.usage.total_tokens,
+                        cachedPromptTokens: (_a = parsed.usage.prompt_tokens_details) === null || _a === void 0 ? void 0 : _a.cached_tokens,
+                    };
+                }
+                const choice = (_b = parsed.choices) === null || _b === void 0 ? void 0 : _b[0];
+                if (!choice)
+                    return;
+                const delta = choice.delta;
+                if (delta.reasoning_content) {
+                    yield { type: 'reasoning', delta: delta.reasoning_content };
+                }
+                if (delta.content) {
+                    yield { type: 'content', delta: delta.content };
+                }
+                if (choice.finish_reason) {
+                    finishReason = choice.finish_reason;
+                    sawFinishReason = true;
+                }
+            }
             try {
                 while (true) {
                     const { done, value } = yield __await(reader.read());
@@ -149,41 +206,31 @@ class BailianProvider {
                     const lines = buffer.split('\n');
                     buffer = (_a = lines.pop()) !== null && _a !== void 0 ? _a : '';
                     for (const line of lines) {
-                        const trimmed = line.trim();
-                        if (!trimmed || !trimmed.startsWith('data: '))
-                            continue;
-                        const data = trimmed.slice(6);
-                        if (data === '[DONE]')
-                            continue;
-                        let parsed;
-                        try {
-                            parsed = JSON.parse(data);
-                        }
-                        catch (_d) {
-                            continue;
-                        }
-                        // 从任意 chunk 中收集 usage（包括 choices 为空的独立 usage chunk）
-                        if (parsed.usage) {
-                            streamUsage = {
-                                promptTokens: parsed.usage.prompt_tokens,
-                                completionTokens: parsed.usage.completion_tokens,
-                                totalTokens: parsed.usage.total_tokens,
-                                cachedPromptTokens: (_b = parsed.usage.prompt_tokens_details) === null || _b === void 0 ? void 0 : _b.cached_tokens,
-                            };
+                        for (const chunk of processSseLine(line)) {
+                            yield yield __await(chunk);
                         }
-                        const choice = (_c = parsed.choices) === null || _c === void 0 ? void 0 : _c[0];
-                        if (!choice)
-                            continue;
-                        const delta = choice.delta;
-                        if (delta.reasoning_content) {
-                            yield yield __await({ type: 'reasoning', delta: delta.reasoning_content });
-                        }
-                        if (delta.content) {
-                            yield yield __await({ type: 'content', delta: delta.content });
+                    }
+                }
+                // TextDecoder 的流式模式可能仍持有不完整的多字节字符；EOF 时必须 flush。
+                buffer += decoder.decode();
+                if (buffer) {
+                    const finalLines = buffer.split('\n');
+                    for (const line of finalLines) {
+                        for (const chunk of processSseLine(line)) {
+                            yield yield __await(chunk);
                         }
                     }
                 }
-                yield yield __await({ type: 'finish', usage: streamUsage });
+                if (!sawDone && !sawFinishReason) {
+                    throw new Error(`[${this.name}] SSE stream ended before [DONE] or finish_reason`);
+                }
+                yield yield __await({
+                    type: 'finish',
+                    finishReason,
+                    sawDone,
+                    sawFinishReason,
+                    usage: streamUsage,
+                });
             }
             finally {
                 reader.releaseLock();
@@ -212,6 +259,7 @@ class BailianProvider {
 }
 exports.BailianProvider = BailianProvider;
 BailianProvider.SUPPORTED_MODELS = new Set([
+    'qwen3.8-flash',
     'qwen3.7-max',
     'qwen3.7-flash',
     'qwen-plus',
@@ -219,12 +267,14 @@ BailianProvider.SUPPORTED_MODELS = new Set([
     'qwq-plus',
     'deepseek-v4-pro',
     'deepseek-v4-flash',
+    'deepseek-v4.1-flash',
     'kimi-k2.6',
     'glm-5.1',
     'qwen-vl-plus',
 ]);
 BailianProvider.MODEL_CAPABILITIES = {
     'qwen-max': { jsonMode: true, reasoningEffort: false },
+    'qwen3.8-flash': { jsonMode: true, reasoningEffort: true },
     'qwen3.7-max': { jsonMode: true, reasoningEffort: true },
     'qwen3.7-flash': { jsonMode: true, reasoningEffort: true },
     'qwen-plus': { jsonMode: true, reasoningEffort: true },
@@ -234,6 +284,7 @@ BailianProvider.MODEL_CAPABILITIES = {
     // 通过 adaptRequest 映射为 extra_body.enable_thinking（与 Qwen 同一分支）。
     'deepseek-v4-pro': { jsonMode: true, reasoningEffort: true },
     'deepseek-v4-flash': { jsonMode: true, reasoningEffort: true },
+    'deepseek-v4.1-flash': { jsonMode: true, reasoningEffort: true },
     'kimi-k2.6': { jsonMode: true, reasoningEffort: true },
     'glm-5.1': { jsonMode: false, reasoningEffort: false },
     'qwen-vl-plus': { jsonMode: true, reasoningEffort: false },

package/dist/llm_provider/models.d.ts

@@ -1,7 +1,7 @@
 /**
  * 支持的模型枚举。项目层通过此枚举选择模型，Axiom 内部路由到对应 Provider。
  */
-export type Model = 'qwen3.7-max' | 'qwen3.7-flash' | 'qwen-plus' | 'qwen-turbo' | 'qwq-plus' | 'deepseek-v4-pro' | 'deepseek-v4-flash' | 'kimi-k2.6' | 'glm-5.1' | 'qwen-vl-plus';
+export type Model = 'qwen3.8-flash' | 'qwen3.7-max' | 'qwen3.7-flash' | 'qwen-plus' | 'qwen-turbo' | 'qwq-plus' | 'deepseek-v4-pro' | 'deepseek-v4-flash' | 'deepseek-v4.1-flash' | 'kimi-k2.6' | 'glm-5.1' | 'qwen-vl-plus';
 /** 模型到 Provider 的映射配置 */
 export interface ModelConfig {
     readonly model: string;

package/dist/llm_provider/models.js

@@ -6,6 +6,7 @@ exports.MODEL_REGISTRY = void 0;
  * 当首选 Provider 失败时，Predictor 按此表顺序尝试下一个。
  */
 exports.MODEL_REGISTRY = {
+    'qwen3.8-flash': [{ model: 'qwen3.8-flash', provider: 'bailian' }],
     'qwen3.7-max': [{ model: 'qwen3.7-max', provider: 'bailian' }],
     'qwen3.7-flash': [{ model: 'qwen3.7-flash', provider: 'bailian' }],
     'qwen-plus': [{ model: 'qwen-plus', provider: 'bailian' }],
@@ -13,6 +14,7 @@ exports.MODEL_REGISTRY = {
     'qwq-plus': [{ model: 'qwq-plus', provider: 'bailian' }],
     'deepseek-v4-pro': [{ model: 'deepseek-v4-pro', provider: 'bailian' }],
     'deepseek-v4-flash': [{ model: 'deepseek-v4-flash', provider: 'bailian' }],
+    'deepseek-v4.1-flash': [{ model: 'deepseek-v4.1-flash', provider: 'bailian' }],
     'kimi-k2.6': [{ model: 'kimi-k2.6', provider: 'bailian' }],
     'glm-5.1': [{ model: 'glm-5.1', provider: 'bailian' }],
     'qwen-vl-plus': [{ model: 'qwen-vl-plus', provider: 'bailian' }],

package/dist/llm_provider/types.d.ts

@@ -40,6 +40,8 @@ export interface LLMResponse {
     readonly reasoningContent?: string;
     readonly usage?: TokenUsage;
     readonly model: string;
+    /** 上游返回的生成结束原因，例如 `stop`、`length` 或 `content_filter`；Provider 未提供时为 undefined。 */
+    readonly finishReason?: string;
 }
 /**
  * 流式输出片段。迭代器每次 yield 一个 chunk。
@@ -52,7 +54,13 @@ export type StreamChunk = {
     readonly delta: string;
 } | {
     readonly type: 'finish';
-    usage?: LLMResponse['usage'];
+    /** 上游返回的生成结束原因，例如 `stop`、`length` 或 `content_filter`。 */
+    readonly finishReason?: string;
+    /** SSE 流是否收到显式的 `data: [DONE]` 终止标记；未提供表示 Provider 不支持该观测。 */
+    readonly sawDone?: boolean;
+    /** SSE 流是否收到非空的 `finish_reason`；未提供表示 Provider 不支持该观测。 */
+    readonly sawFinishReason?: boolean;
+    readonly usage?: LLMResponse['usage'];
 };
 /**
  * Provider 配置。每个 Provider 实例需要一组连接参数。

package/dist/predict/predict.d.ts

@@ -26,6 +26,7 @@ export declare class LLM {
      * @param config - 预测配置（模型、prompt、温度等）
      * @returns 模型生成的完整响应及 token 消耗统计
      * @throws Error - Provider 未配置、模型不存在、或所有 Provider 均失败时抛出
+     * @throws Error - `responseFormat: 'json'` 时输出为空或无法解析为 JSON；message 含 model、finish_reason、原文
      */
     static predict(config: PredictConfig & {
         responseFormat: 'json';
@@ -48,6 +49,7 @@ export declare class LLM {
      * @param config - 预测配置（不含 prompt，因为由 messages 提供）
      * @returns 模型生成的完整响应及 token 消耗统计
      * @throws Error - Provider 未配置、模型不存在、或所有 Provider 均失败时抛出
+     * @throws Error - `responseFormat: 'json'` 时输出为空或无法解析为 JSON；message 含 model、finish_reason、原文
      */
     static predictWithMessages(messages: ReadonlyArray<Message>, config: PredictWithMessagesConfig & {
         responseFormat: 'json';

package/dist/predict/predict.js

@@ -69,16 +69,24 @@ function toLLMRequest(config, messages) {
     };
 }
 function unwrapResponse(response, responseFormat) {
-    var _a;
+    var _a, _b;
     const usage = response.usage;
     if (responseFormat === 'json') {
+        // 原文与 finish_reason 写进 message：调用方只 catch 到异常时也能从日志判断是截断（length）还是模型输出非法 JSON（stop）
+        const context = `model=${response.model}, finish_reason=${(_a = response.finishReason) !== null && _a !== void 0 ? _a : 'unknown'}`;
         if (!response.content) {
-            throw new Error('Empty response content when responseFormat is json');
+            throw new Error(`Empty response content when responseFormat is json (${context})`);
+        }
+        try {
+            return { content: JSON.parse(response.content), usage };
+        }
+        catch (cause) {
+            const message = cause instanceof Error ? cause.message : String(cause);
+            throw new Error(`JSON parse failed: ${message} (${context}, raw_length=${response.content.length})\nraw: ${response.content}`);
         }
-        return { content: JSON.parse(response.content), usage };
     }
     // 默认按 text 返回（包括未设置 responseFormat 的情况）
-    return { content: (_a = response.content) !== null && _a !== void 0 ? _a : '', usage };
+    return { content: (_b = response.content) !== null && _b !== void 0 ? _b : '', usage };
 }
 /**
  * LLM 静态类。封装 Provider 连接、请求组装、故障转移和返回解析。

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@keo-ai/axiom",
-  "version": "0.2.8",
+  "version": "0.2.10",
   "description": "基于 LLM 的预测与推理库，支持多 Provider 切换",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",

package/dist/predict/providers/bailian.d.ts

@@ -1,27 +0,0 @@
-import type { LLMProvider } from '../llm';
-import type { LLMRequest, LLMResponse, ProviderConfig, StreamChunk } from '../types';
-/**
- * 阿里云百炼（兼容 OpenAI 协议）Provider 实现。
- * 支持非流式调用和流式 SSE 输出。
- */
-export declare class BailianProvider implements LLMProvider {
-    readonly config: ProviderConfig;
-    readonly name = "bailian";
-    constructor(config: ProviderConfig);
-    supports(model: string): boolean;
-    /**
-     * 发送非流式 chat completion 请求到百炼服务。
-     * @param request - 标准化 LLM 请求
-     * @returns 解析后的模型响应（含 content、usage、model）
-     * @throws Error - 网络异常时抛 `[bailian] ...`；HTTP 非 2xx 时抛 `[bailian] HTTP {status}: ...`
-     */
-    generate(request: LLMRequest): Promise<LLMResponse>;
-    /**
-     * 发送流式 chat completion 请求，解析 SSE 响应逐块返回。
-     * @param request - 标准化 LLM 请求
-     * @yields 内容片段（`content`）或结束标记（`finish`）
-     * @throws Error - 网络异常或 HTTP 错误时抛出
-     */
-    stream(request: LLMRequest): AsyncGenerator<StreamChunk, void, unknown>;
-    private buildRequestBody;
-}

package/dist/predict/providers/bailian.js

@@ -1,185 +0,0 @@
-"use strict";
-var __awaiter = (this && this.__awaiter) || function (thisArg, _arguments, P, generator) {
-    function adopt(value) { return value instanceof P ? value : new P(function (resolve) { resolve(value); }); }
-    return new (P || (P = Promise))(function (resolve, reject) {
-        function fulfilled(value) { try { step(generator.next(value)); } catch (e) { reject(e); } }
-        function rejected(value) { try { step(generator["throw"](value)); } catch (e) { reject(e); } }
-        function step(result) { result.done ? resolve(result.value) : adopt(result.value).then(fulfilled, rejected); }
-        step((generator = generator.apply(thisArg, _arguments || [])).next());
-    });
-};
-var __await = (this && this.__await) || function (v) { return this instanceof __await ? (this.v = v, this) : new __await(v); }
-var __asyncGenerator = (this && this.__asyncGenerator) || function (thisArg, _arguments, generator) {
-    if (!Symbol.asyncIterator) throw new TypeError("Symbol.asyncIterator is not defined.");
-    var g = generator.apply(thisArg, _arguments || []), i, q = [];
-    return i = Object.create((typeof AsyncIterator === "function" ? AsyncIterator : Object).prototype), verb("next"), verb("throw"), verb("return", awaitReturn), i[Symbol.asyncIterator] = function () { return this; }, i;
-    function awaitReturn(f) { return function (v) { return Promise.resolve(v).then(f, reject); }; }
-    function verb(n, f) { if (g[n]) { i[n] = function (v) { return new Promise(function (a, b) { q.push([n, v, a, b]) > 1 || resume(n, v); }); }; if (f) i[n] = f(i[n]); } }
-    function resume(n, v) { try { step(g[n](v)); } catch (e) { settle(q[0][3], e); } }
-    function step(r) { r.value instanceof __await ? Promise.resolve(r.value.v).then(fulfill, reject) : settle(q[0][2], r); }
-    function fulfill(value) { resume("next", value); }
-    function reject(value) { resume("throw", value); }
-    function settle(f, v) { if (f(v), q.shift(), q.length) resume(q[0][0], q[0][1]); }
-};
-Object.defineProperty(exports, "__esModule", { value: true });
-exports.BailianProvider = void 0;
-const models_1 = require("../models");
-/**
- * 阿里云百炼（兼容 OpenAI 协议）Provider 实现。
- * 支持非流式调用和流式 SSE 输出。
- */
-class BailianProvider {
-    constructor(config) {
-        this.config = config;
-        this.name = 'bailian';
-    }
-    supports(model) {
-        return Object.entries(models_1.MODEL_REGISTRY).some(([m, configs]) => m === model && configs.some((c) => c.provider === this.name));
-    }
-    /**
-     * 发送非流式 chat completion 请求到百炼服务。
-     * @param request - 标准化 LLM 请求
-     * @returns 解析后的模型响应（含 content、usage、model）
-     * @throws Error - 网络异常时抛 `[bailian] ...`；HTTP 非 2xx 时抛 `[bailian] HTTP {status}: ...`
-     */
-    generate(request) {
-        return __awaiter(this, void 0, void 0, function* () {
-            const url = `${this.config.baseUrl}/chat/completions`;
-            const body = this.buildRequestBody(request);
-            let response;
-            try {
-                response = yield fetch(url, {
-                    method: 'POST',
-                    headers: {
-                        'Content-Type': 'application/json',
-                        Authorization: `Bearer ${this.config.apiKey}`,
-                    },
-                    body: JSON.stringify(Object.assign(Object.assign({}, body), { stream: false })),
-                });
-            }
-            catch (cause) {
-                throw new Error(`[${this.name}] ${cause instanceof Error ? cause.message : String(cause)}`);
-            }
-            if (!response.ok) {
-                const text = yield response.text();
-                throw new Error(`[${this.name}] HTTP ${response.status}: ${text}`);
-            }
-            const data = (yield response.json());
-            const choice = data.choices[0];
-            if (!choice) {
-                throw new Error(`[${this.name}] No choice in response`);
-            }
-            return {
-                content: choice.message.content,
-                usage: data.usage
-                    ? {
-                        promptTokens: data.usage.prompt_tokens,
-                        completionTokens: data.usage.completion_tokens,
-                        totalTokens: data.usage.total_tokens,
-                    }
-                    : undefined,
-                model: data.model,
-            };
-        });
-    }
-    /**
-     * 发送流式 chat completion 请求，解析 SSE 响应逐块返回。
-     * @param request - 标准化 LLM 请求
-     * @yields 内容片段（`content`）或结束标记（`finish`）
-     * @throws Error - 网络异常或 HTTP 错误时抛出
-     */
-    stream(request) {
-        return __asyncGenerator(this, arguments, function* stream_1() {
-            var _a, _b;
-            const url = `${this.config.baseUrl}/chat/completions`;
-            const body = this.buildRequestBody(request);
-            let response;
-            try {
-                response = yield __await(fetch(url, {
-                    method: 'POST',
-                    headers: {
-                        'Content-Type': 'application/json',
-                        Authorization: `Bearer ${this.config.apiKey}`,
-                    },
-                    body: JSON.stringify(Object.assign(Object.assign({}, body), { stream: true })),
-                }));
-            }
-            catch (cause) {
-                throw new Error(`[${this.name}] ${cause instanceof Error ? cause.message : String(cause)}`);
-            }
-            if (!response.ok) {
-                const text = yield __await(response.text());
-                throw new Error(`[${this.name}] HTTP ${response.status}: ${text}`);
-            }
-            if (!response.body) {
-                throw new Error(`[${this.name}] Response body is null`);
-            }
-            const reader = response.body.getReader();
-            const decoder = new TextDecoder();
-            let buffer = '';
-            try {
-                while (true) {
-                    const { done, value } = yield __await(reader.read());
-                    if (done)
-                        break;
-                    buffer += decoder.decode(value, { stream: true });
-                    const lines = buffer.split('\n');
-                    buffer = (_a = lines.pop()) !== null && _a !== void 0 ? _a : '';
-                    for (const line of lines) {
-                        const trimmed = line.trim();
-                        if (!trimmed || !trimmed.startsWith('data: '))
-                            continue;
-                        const data = trimmed.slice(6);
-                        if (data === '[DONE]')
-                            continue;
-                        let parsed;
-                        try {
-                            parsed = JSON.parse(data);
-                        }
-                        catch (_c) {
-                            continue;
-                        }
-                        const choice = (_b = parsed.choices) === null || _b === void 0 ? void 0 : _b[0];
-                        if (!choice)
-                            continue;
-                        const delta = choice.delta;
-                        if (delta.content) {
-                            yield yield __await({ type: 'content', delta: delta.content });
-                        }
-                        if (choice.finish_reason && parsed.usage) {
-                            yield yield __await({
-                                type: 'finish',
-                                usage: {
-                                    promptTokens: parsed.usage.prompt_tokens,
-                                    completionTokens: parsed.usage.completion_tokens,
-                                    totalTokens: parsed.usage.total_tokens,
-                                },
-                            });
-                        }
-                    }
-                }
-                yield yield __await({ type: 'finish' });
-            }
-            finally {
-                reader.releaseLock();
-            }
-        });
-    }
-    buildRequestBody(request) {
-        var _a;
-        return {
-            model: (_a = request.model) !== null && _a !== void 0 ? _a : this.config.defaultModel,
-            messages: request.messages.map((m) => ({
-                role: m.role,
-                content: m.content,
-            })),
-            temperature: request.temperature,
-            max_tokens: request.maxTokens,
-            top_p: request.topP,
-            response_format: request.responseFormat
-                ? { type: request.responseFormat === 'json' ? 'json_object' : 'text' }
-                : undefined,
-        };
-    }
-}
-exports.BailianProvider = BailianProvider;