/** * OpenAI Provider Adapter(P3) * * OpenAI Chat Completions API(/v1/chat/completions),支持 Tool Calling、 * 流式输出、多模态图片、o 系列推理模型的 reasoning_effort 参数。 * * 与 DeepSeek 适配器的关键差异: * - o 系列 / gpt-5 系列模型使用 max_completion_tokens(非 max_tokens) * - Thinking 模式通过顶层 reasoning_effort 参数(o 系列模型) * - 模型列表从 /v1/models 动态获取 * * @see apis 官方文档 https://platform.openai.com/docs/api-reference/chat */ import { BaseAdapter } from './base-adapter'; import type { MetonaRequest, MetonaResponse, MetonaStreamEvent } from '../types'; import { MetonaFinishReason } from '../types'; import type { MetonaModelInfo } from '../types/metona-adapter'; import { buildOpenAICompatibleMessages, buildOpenAICompatibleTools } from './shared/openai-format'; import { parseSSEStream, parseOpenAICompatibleResponse } from './shared/sse-stream'; export class OpenAIAdapter extends BaseAdapter { override readonly providerId: string = 'openai'; readonly supportedModels = ['gpt-4o', 'gpt-4o-mini', 'gpt-4.1', 'o3-mini']; readonly supportsToolCalling = true; readonly supportsThinking = true; private static readonly MODEL_INFO: Record = { 'gpt-4o': { id: 'gpt-4o', name: 'GPT-4o', contextWindow: 128_000, maxOutputTokens: 16_384, supportsToolCalling: true, supportsThinking: false, description: 'OpenAI 旗舰多模态模型,128K 上下文', }, 'gpt-4o-mini': { id: 'gpt-4o-mini', name: 'GPT-4o mini', contextWindow: 128_000, maxOutputTokens: 16_384, supportsToolCalling: true, supportsThinking: false, description: 'OpenAI 高性价比模型,128K 上下文', }, 'gpt-4.1': { id: 'gpt-4.1', name: 'GPT-4.1', contextWindow: 1_000_000, maxOutputTokens: 32_768, supportsToolCalling: true, supportsThinking: false, description: 'OpenAI 长上下文模型,1M 上下文', }, 'o3-mini': { id: 'o3-mini', name: 'o3-mini', contextWindow: 200_000, maxOutputTokens: 100_000, supportsToolCalling: true, supportsThinking: true, description: 'OpenAI 推理模型,支持 reasoning_effort', }, }; // ===== POST /v1/chat/completions(非流式) ===== async send(request: MetonaRequest): Promise { const body = this.toNativeRequest(request, false); const response = await this.fetchWithTimeout( `${this.config.baseURL}/chat/completions`, { method: 'POST', headers: { 'Content-Type': 'application/json', Authorization: `Bearer ${this.config.apiKey}`, ...this.config.headers, }, body: JSON.stringify(body), }, this.config.timeoutMs ?? 120_000, ); if (!response.ok) { await this.throwHttpError(response, 'OpenAI API error'); } const data = (await response.json()) as Record; const parsed = parseOpenAICompatibleResponse(data); return { meta: { requestId: request.meta.requestId, provider: this.providerId, model: (data.model as string) ?? this.config.defaultModel, latencyMs: 0, timestamp: Date.now(), }, content: parsed.content, reasoningContent: parsed.reasoningContent, toolCalls: parsed.toolCalls, usage: parsed.usage, finishReason: parsed.finishReason as MetonaFinishReason, }; } // ===== POST /v1/chat/completions(流式) ===== async *sendStream(request: MetonaRequest): AsyncIterable { const body = this.toNativeRequest(request, true); const response = await this.fetchWithTimeout( `${this.config.baseURL}/chat/completions`, { method: 'POST', headers: { 'Content-Type': 'application/json', Authorization: `Bearer ${this.config.apiKey}`, ...this.config.headers, }, body: JSON.stringify(body), }, this.config.timeoutMs ?? 300_000, ); if (!response.ok || !response.body) { await this.throwHttpError(response, 'OpenAI stream error'); } yield* parseSSEStream( // 非空断言:上方 if 已确保 response.body 不为 null response.body!, request.meta.requestId, request.meta.sessionId, request.meta.iteration, ); } // ===== GET /v1/models ===== override async listModels(): Promise { try { const response = await fetch(`${this.config.baseURL}/models`, { headers: { Authorization: `Bearer ${this.config.apiKey}` }, signal: AbortSignal.timeout(10_000), }); if (response.ok) { const data = (await response.json()) as { data?: Array<{ id: string }> }; if (data.data?.length) { return data.data.map((m) => OpenAIAdapter.MODEL_INFO[m.id] ?? { id: m.id }); } } } catch { // API 不可用时降级 } return this.supportedModels.map((id) => OpenAIAdapter.MODEL_INFO[id] ?? { id }); } override getContextWindow(): number { if (typeof this.config.contextWindow === 'number' && this.config.contextWindow > 0) { return this.config.contextWindow; } const modelInfo = OpenAIAdapter.MODEL_INFO[this.config.defaultModel]; return modelInfo?.contextWindow ?? 128_000; } // ========== 私有方法 ========== /** * 构建 OpenAI 原生请求体 * * OpenAI 特有处理: * - 多模态图片:user 消息 images[] → content 数组 * - o 系列(o1/o3/o4)与 gpt-5 系列使用 max_completion_tokens + reasoning_effort * - 思考模式下 temperature 被部分推理模型拒绝,不传 */ private toNativeRequest(request: MetonaRequest, stream: boolean): Record { // 推理模型检测(o 系列使用新参数名) const model = this.config.defaultModel; const isReasoningModel = /^(o\d|gpt-5)/.test(model); // 推理模型不支持图片输入 — 前置校验(转换在共享层,此处仅拦截) if (isReasoningModel) { const hasImages = request.messages.some((m) => m.images?.length); if (hasImages) { throw new Error(`Model "${model}" does not support image inputs`); } } // v0.6.2: images 处理收敛至共享层(原索引对齐循环在孤立 tool 过滤后会错位) const messages = buildOpenAICompatibleMessages(request, true); const tools = buildOpenAICompatibleTools(request.tools); const body: Record = { model, messages, stream, }; // Token 上限参数:o 系列/gpt-5 使用 max_completion_tokens // v0.5.3: 按模型上限钳制(gpt-4o 16384 / gpt-4.1 32768 / o3-mini 100000)— // 引擎默认 63488 超过 gpt-4o/gpt-4.1 上限时 API 直接 400 const oaMaxOutput = OpenAIAdapter.MODEL_INFO[model]?.maxOutputTokens ?? 128_000; const oaMaxTokens = Math.min( request.params.maxTokens ?? (isReasoningModel ? 32_768 : oaMaxOutput), oaMaxOutput, ); if (oaMaxTokens) { if (isReasoningModel) { body.max_completion_tokens = oaMaxTokens; } else { body.max_tokens = oaMaxTokens; } } if (stream) { body.stream_options = { include_usage: true }; } if (tools) { body.tools = tools; } // Thinking 模式:推理模型映射 reasoning_effort;非推理模型忽略 if (request.params.thinkingEnabled && isReasoningModel) { const effortMap: Record = { low: 'low', medium: 'medium', high: 'high', max: 'high', }; body.reasoning_effort = effortMap[request.params.thinkingEffort ?? 'high'] ?? 'high'; } else if (!isReasoningModel) { // 非推理模型使用温度控制 body.temperature = request.params.temperature; } // 停止序列 if (request.params.stopSequences?.length) { body.stop = request.params.stopSequences; } return body; } }