feat: 添加 MiMo 模型 thinking mode 自动检测与兼容

isOpenAIThinkingEnabled() 现在自动检测模型名包含 "mimo" 的模型
(与 DeepSeek 并列),因为 MiMo 同样使用 reasoning_content 字段
且支持 thinking mode。

buildOpenAIRequestBody() 在 chat_template_kwargs 中同时发送
thinking: true 和 enable_thinking: true,兼容 DeepSeek 自托管和
MiMo 的 thinking 启用格式。

已有 reasoning_content 回传逻辑(openaiConvertMessages.ts)和流
解析逻辑(openaiStreamAdapter.ts)无需修改,MiMo 与 DeepSeek 共用
相同的 reasoning_content 字段协议。

Co-Authored-By: deepseek-v4-pro[1m] <deepseek-ai@claude-code-best.win>
This commit is contained in:
cepvor
2026-05-14 16:47:12 +08:00
parent 80b46d2221
commit 833181e025
2 changed files with 34 additions and 14 deletions

View File

@@ -7,11 +7,11 @@ import type { ChatCompletionCreateParamsStreaming } from 'openai/resources/chat/
import { isEnvTruthy, isEnvDefinedFalsy } from '../../../utils/envUtils.js'
/**
* Detect whether DeepSeek-style thinking mode should be enabled.
* Detect whether thinking mode should be enabled for this model.
*
* Enabled when:
* 1. OPENAI_ENABLE_THINKING=1 is set (explicit enable), OR
* 2. Model name contains "deepseek-reasoner" OR "DeepSeek-V3.2" (auto-detect, case-insensitive)
* 2. Model name contains "deepseek" or "mimo" (auto-detect, case-insensitive)
*
* Disabled when:
* - OPENAI_ENABLE_THINKING=0/false/no/off is explicitly set (overrides model detection)
@@ -23,9 +23,9 @@ export function isOpenAIThinkingEnabled(model: string): boolean {
if (isEnvDefinedFalsy(process.env.OPENAI_ENABLE_THINKING)) return false
// Explicit enable
if (isEnvTruthy(process.env.OPENAI_ENABLE_THINKING)) return true
// Auto-detect from model name (all DeepSeek models support thinking mode)
// Auto-detect from model name (DeepSeek and MiMo models support thinking mode)
const modelLower = model.toLowerCase()
return modelLower.includes('deepseek')
return modelLower.includes('deepseek') || modelLower.includes('mimo')
}
/**
@@ -58,12 +58,12 @@ export function resolveOpenAIMaxTokens(
* Build the request body for OpenAI chat.completions.create().
* Extracted for testability — the thinking mode params are injected here.
*
* DeepSeek thinking mode: inject thinking params via request body.
* Two formats are added simultaneously to support different deployments:
* - Official DeepSeek API: `thinking: { type: 'enabled' }`
* - Self-hosted DeepSeek-V3.2: `enable_thinking: true` + `chat_template_kwargs: { thinking: true }`
* Three thinking-mode formats are sent simultaneously; each endpoint uses the
* format it recognizes and ignores the others:
* - Official DeepSeek API: `thinking: { type: 'enabled' }`
* - Self-hosted DeepSeek: `enable_thinking: true` + `chat_template_kwargs: { thinking: true }`
* - MiMo (Xiaomi): `chat_template_kwargs: { enable_thinking: true }`
* OpenAI SDK passes unknown keys through to the HTTP body.
* Each endpoint will use the format it recognizes and ignore the others.
*/
export function buildOpenAIRequestBody(params: {
model: string
@@ -76,7 +76,7 @@ export function buildOpenAIRequestBody(params: {
}): ChatCompletionCreateParamsStreaming & {
thinking?: { type: string }
enable_thinking?: boolean
chat_template_kwargs?: { thinking: boolean }
chat_template_kwargs?: { thinking: boolean; enable_thinking: boolean }
} {
const {
model,
@@ -97,14 +97,15 @@ export function buildOpenAIRequestBody(params: {
}),
stream: true,
stream_options: { include_usage: true },
// DeepSeek thinking mode: enable chain-of-thought output.
// When active, temperature/top_p/presence_penalty/frequency_penalty are ignored by DeepSeek.
// Enable chain-of-thought output for DeepSeek and MiMo models.
// When active, temperature/top_p/presence_penalty/frequency_penalty are ignored.
...(enableThinking && {
// Official DeepSeek API format
thinking: { type: 'enabled' },
// Self-hosted DeepSeek-V3.2 format
enable_thinking: true,
chat_template_kwargs: { thinking: true },
// Both DeepSeek self-hosted and MiMo formats in chat_template_kwargs
chat_template_kwargs: { thinking: true, enable_thinking: true },
}),
// Only send temperature when thinking mode is off (DeepSeek ignores it anyway,
// but other providers may respect it)