Feature/mimo vllm asr

[yuanyp8]

User description

摘要

Fixes #。

说明这个 PR 解决的单一目标。标题必须使用 type(scope): 中文目标，例如 docs(infra): 补齐 Windows 构建说明。

修复 / 新增 / 改进

兼容

• 不包含：
• 对现有用户 / 本地环境 / 构建流程的影响：

测试计划

• 命令：
• 结果：
• 证据路径：

---

PR Type

Enhancement, Documentation

---

Description

• Adapt MiMo ASR provider to vLLM-Omni API (audio_url, empty API key allowed)

• Add URL normalization for bare IP/missing path

• Add detailed logging for each chunk and request

• Document setup guide for self-hosted deployment

---

Diagram Walkthrough

flowchart LR
  PCM["PCM Audio"] --> chunk["Split into chunks (30s max)"]
  chunk --> transcribe_chunk["transcribe_chunk()"]
  transcribe_chunk --> POST["POST /v1/chat/completions"]
  POST -- "audio_url (base64 WAV), text prompt" --> vLLM["vLLM-Omni server"]
  vLLM -- "JSON choice text" --> extract["extract_mimo_text()"]
  extract --> join["join_transcript_chunks()"]
  join --> RawTranscript["RawTranscript"]
  subgraph Config
    base_url["Endpoint (IP/host)"] --> URL["normalize to /v1/chat/completions"]
    api_key["API Key (optional)"] --> auth["Authorization header only if key non-empty"]
    model["Model name"] --> body["JSON body"]
  end
  body --> POST
  URL --> POST
  auth --> POST

Loading

File Walkthrough

	Relevant files
Enhancement	mimo.rs Adapt MiMo ASR to vLLM-Omni API                                                    openless-all/app/src-tauri/src/asr/mimo.rs Switch from input_audio to audio_url payload format for vLLM-Omni compatibility Allow empty API key; only send Authorization header when key is set Add URL normalization (auto-prepend http:// and /v1/chat/completions path) Add detailed logging for each API call (chunk index, elapsed, status) Update tests to reflect new request format and add test for empty API key +151/-37
Documentation	mimo-vllm-asr.md Add self-hosted ASR setup guide                                                    docs/mimo-vllm-asr.md Document self-hosted MiMo-V2.5-ASR setup instructions Explain configuration fields (Endpoint, Model, API Key) List key changes: URL normalization, API key optional, new request body Include troubleshooting tips (HTTP error codes) and verification commands +93/-0

---

[github-actions]

PR Reviewer Guide 🔍

(Review updated until commit 27c51f3)

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 2 🔵🔵⚪⚪⚪

🧪 PR contains tests

🔒 No security concerns identified

⚡ No major issues detected

[github-actions]

Persistent review updated to latest commit e4c539a

[github-actions]

Persistent review updated to latest commit 27c51f3

[appergb]

Review note: this should not be merged as-is without a product decision and full platform CI. The implementation changes the existing MiMo provider from the official input_audio shape to vLLM-Omni audio_url, and also permits empty API keys. That is useful for self-hosted vLLM, but it may break users of the official Xiaomi MiMo endpoint. Safer options: add an explicit self-hosted mode/provider, or keep the official request shape and gate vLLM-specific behavior by config.