Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.
Python asraudiochineseemotion-recognitionfunasrmcp-servermultilingual-asropenai-compatible-apiparaformerpunctuationpytorchreal-time-asrspeaker-diarizationspeech-recognitionspeech-to-textstreaming-asrtranscriptionvllmvoice-activity-detectionwhisper-alternative
docs: add GLM-ASR vLLM benchmark results to vllm_guide.md
游雁674ba5e53e2c
diff --git a/docs/vllm_guide.md b/docs/vllm_guide.mdindex 7e0c267de..dbff38109 100644--- a/docs/vllm_guide.md+++ b/docs/vllm_guide.md@@ -25,7 +25,7 @@ FunASR 集成 vLLM 高吞吐推理引擎,用于加速 LLM-based ASR 模型的|------|-----------|------|| **FunASRNano** | ✓ | Qwen3-0.6B LLM,支持离线和流式 || **LLMASR** | ✓ | Whisper + Qwen/Vicuna/LLaMA |-| **GLMASR** | ✓ | GLM-ASR-Nano |+| **GLMASR** | ✓ | GLM-ASR-Nano (RTFx 263, CER 12.92%) || **QwenAudioWarp** | ✓ | Qwen-Audio || Paraformer | ✗ | 非自回归模型(CIF predictor),无 LLM 解码 || SenseVoice | ✗ | Whisper-like encoder-decoder,非 LLM |@@ -384,6 +384,26 @@ Client Server| vLLM batch generate (all prompts) | LLM 解码加速 ~20x || 去掉 `<think>` tokens | 减少无效生成步骤 |+### GLM-ASR-Nano++| 方法 | RTFx | CER | 加速 |+|------|------|-----|------|+| PyTorch native | 34.4 | 12.94% | 基准 |+| **vLLM (ours)** | **263.2** | **12.92%** | **7.6x** |++```python+from funasr.models.glm_asr.inference_vllm import GLMASRVLLMEngine++engine = GLMASRVLLMEngine.from_pretrained(+ model="zai-org/GLM-ASR-Nano-2512",+ hub="ms",+ gpu_memory_utilization=0.4,+ max_model_len=4096,+)+results = engine.generate(inputs=["audio.wav"])+print(results[0]["text"])+```+### WebSocket 实时服务| 指标 | 数值 |