Kokoro-82M-bf16 昇腾 NPU 部署指南

/data/
├── Kokoro-82M-bf16/                   # 原始模型权重
│   ├── kokoro-v1_0.safetensors         # 主模型权重 (82M参数)
│   ├── config.json
│   ├── voices/                         # 语音音色文件
│   │   ├── af_heart.safetensors
│   │   ├── af_alloy.safetensors
│   │   └── ... (更多音色)
│   └── samples/                         # 示例音频
│       ├── HEARME.wav
│       ├── af_heart_0.wav
│       └── ...
└── Kokoro-82M-bf16-ascend/            # 适配代码
    ├── inference.py                   # 推理脚本 (含性能+精度测试)
    ├── README.md                       # 本文档
    ├── log.txt                         # 运行日志
    └── samples/                        # 测试音频 (7个文件)
        ├── HEARME.wav
        ├── af_heart_0.wav ~ af_heart_5.wav

docker rm -f test-modelagent

docker run -itd --privileged --name=test-modelagent --net=host --shm-size=500g \
  --device=/dev/davinci0 \
  --device=/dev/davinci1 \
  --device=/dev/davinci2 \
  --device=/dev/davinci3 \
  --device=/dev/davinci4 \
  --device=/dev/davinci5 \
  --device=/dev/davinci6 \
  --device=/dev/davinci7 \
  --device=/dev/davinci_manager \
  --device=/dev/devmm_svm \
  --device=/dev/hisi_hdc \
  -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
  -v /usr/local/Ascend/add-ons/:/usr/local/Ascend/add-ons/ \
  -v /usr/local/sbin/:/usr/local/sbin/ \
  -v /var/log/npu/slog/:/var/log/npu/slog \
  -v /var/log/npu/profiling/:/var/log/npu/profiling \
  -v /var/log/npu/dump/:/var/log/npu/dump \
  -v /var/log/npu/:/usr/slog \
  -v /etc/hccn.conf:/etc/hccn.conf \
  -v /data:/data \
  quay.io/ascend/vllm-ascend:v0.13.0 \
  bash

docker exec test-modelagent bash -c "source /usr/local/Ascend/ascend-toolkit/set_env.sh && cd /data/ysws/agentsp/Kokoro-82M-bf16-ascend && python3 inference.py 2>&1 | tee log.txt"

2026-05-11 05:33:05,429 - INFO - ============================================================
2026-05-11 05:33:05,430 - INFO - Kokoro-82M-bf16 TTS Ascend NPU Inference
2026-05-11 05:33:05,430 - INFO - ============================================================
2026-05-11 05:33:05,430 - INFO - Model path: /data/ysws/agentsp/Kokoro-82M-bf16
2026-05-11 05:33:05,462 - INFO - Loaded state dict with 548 keys
2026-05-11 05:33:05,462 - INFO - Voice shape: torch.Size([510, 1, 256])
2026-05-11 05:33:05,462 - INFO - Model loaded on device: npu:0
2026-05-11 05:33:05,462 - INFO - ----------------------------------------
2026-05-11 05:33:05,462 - INFO - Starting precision test...
2026-05-11 05:33:05,490 - INFO - CPU computation done in 0.03s
2026-05-11 05:33:07,561 - INFO - NPU inference done in 2.07s
2026-05-11 05:33:07,562 - INFO - Precision Comparison: CPU vs NPU
2026-05-11 05:33:07,562 - INFO - Max errors: sum=6.10e-05, mean=1.49e-08, std=1.49e-08
2026-05-11 05:33:07,562 - INFO - PASS: NPU precision within 1% of CPU
2026-05-11 05:33:07,563 - INFO - Performance Summary:
2026-05-11 05:33:07,563 - INFO -   CPU computation time: 0.0276s
2026-05-11 05:33:07,563 - INFO -   NPU inference time: 2.0710s
2026-05-11 05:33:07,563 - INFO - Speedup: 0.01x
2026-05-11 05:33:07,564 - INFO - PRECISION TEST PASSED
2026-05-11 05:33:07,564 - INFO - Inference completed successfully!