❌ C++ 推理服务启动失败 #77
Description
Describe the bug
MiniCPM-o 一键部署脚本
🖥️ 本机 IP: 192.168.50.254
📋 模式: duplex
🔌 端口: 9060
📁 CPP_DIR: /Users/user/test/AI/llama.cpp-omni
📁 MODEL_DIR: /Users/user/test/AI/gguf
[1/7] 检查 Docker...
✅ Docker 已就绪
[2/7] 更新 LiveKit 配置...
✅ LiveKit 配置已更新 (IP: 192.168.50.254)
[3/7] 加载 Docker 镜像...
加载前端镜像...
Loaded image: modelbest-registry.cn-beijing.cr.aliyuncs.com/modelbest/three-o-h-100:kol-external-202602041835-b7c0c5
✅ 前端镜像已加载
加载后端镜像...
Loaded image: modelbest-registry.cn-beijing.cr.aliyuncs.com/modelbest/minicpm-web-backend:release_local-202602041743-4f385b
✅ 后端镜像已加载
[4/7] 启动 Docker 服务...
WARN[0000] /Users/user/test/AI/omni_docker/docker-compose.yml: the attribute version is obsolete, it will be ignored, please remove it to avoid potential confusion
WARN[0000] No services to build
[+] up 7/7
✔ Network omni_docker_minicpmo-net Created 0.0s
✔ Container minicpmo-redis Created 0.1s
✔ Container minicpmo-livekit Created 0.1s
✔ Container minicpmo-backend Created 0.1s
! backend The requested image's platform (linux/amd64) does not match the detected host platform (linux/arm64/v8) and no specific platform was requested 0.0s
✔ Container minicpmo-frontend Created 0.0s
! frontend The requested image's platform (linux/amd64) does not match the detected host platform (linux/arm64/v8) and no specific platform was requested 0.0s
等待服务启动...
WARN[0000] /Users/user/test/AI/omni_docker/docker-compose.yml: the attribute version is obsolete, it will be ignored, please remove it to avoid potential confusion
✅ Docker 服务已启动
WARN[0000] /Users/user/test/AI/omni_docker/docker-compose.yml: the attribute version is obsolete, it will be ignored, please remove it to avoid potential confusion
NAME IMAGE COMMAND SERVICE CREATED STATUS PORTS
minicpmo-backend modelbest-registry.cn-beijing.cr.aliyuncs.com/modelbest/minicpm-web-backend:release_local-202602041743-4f385b "sh -c 'uvicorn main…" backend 11 seconds ago Up 10 seconds (health: starting) 0.0.0.0:8021-8022->8021-8022/tcp, [::]:8021-8022->8021-8022/tcp
minicpmo-frontend modelbest-registry.cn-beijing.cr.aliyuncs.com/modelbest/three-o-h-100:kol-external-202602041835-b7c0c5 "/docker-entrypoint.…" frontend 11 seconds ago Up 10 seconds 0.0.0.0:3000->80/tcp, [::]:3000->80/tcp
minicpmo-livekit livekit/livekit-server:v1.5.3 "/livekit-server --c…" livekit 11 seconds ago Up 11 seconds 0.0.0.0:7882->7882/udp, [::]:7882->7882/udp, 0.0.0.0:7880-7881->7880-7881/tcp, [::]:7880-7881->7880-7881/tcp, 0.0.0.0:50000-50100->50000-50100/udp, [::]:50000-50100->50000-50100/udp
minicpmo-redis redis:7-alpine "docker-entrypoint.s…" redis 11 seconds ago Up 11 seconds 0.0.0.0:6379->6379/tcp, [::]:6379->6379/tcp
[5/7] 检查 Python 依赖...
安装 Python 依赖...
✅ Python 环境已就绪
[6/7] 启动 C++ 推理服务...
等待推理服务启动...
❌ C++ 推理服务启动失败
查看日志: tail -f /tmp/cpp_server.log
[CPP] alloc_compute_meta: CPU compute buffer size = 18.30 MiB
[CPP] init tts....init t2w....ggml_metal_init: allocating
[CPP] ggml_metal_init: found device: Apple M1
[CPP] ggml_metal_init: picking default device: Apple M1
[CPP] ggml_metal_init: use bfloat = true
[CPP] ggml_metal_init: use fusion = true
[CPP] ggml_metal_init: use concurrency = true
[CPP] ggml_metal_init: use graph optimize = true
[CPP] flowGGUFModelLoader: init_backend device=gpu:1, gpu_idx=1, backend=Metal
[CPP] ggml_metal_free: deallocating
[CPP] omni_warmup_ane: starting ANE/CoreML warmup...
[CPP] omni_warmup_ane: ANE/CoreML warmup finished
[CPP] audition_audio_preprocess: Decoded audio - sample_rate=16000, n_samples=96256
[CPP] audition_audio_preprocess: Audio not aligned to 100ms (96256 samples = 6016.0ms), padded with silence to 97600 samples (6100ms)
[CPP] audition_get_mel_filters: Retrieved mel filters from model: 80 mel bins x 201 fft bins
[CPP] audition_audio_preprocess: Using mel filters - n_mel=80, n_fft=201
[CPP] audition_audio_preprocess: Generated 1 mel spectrogram chunk(s)
[CPP] audition_audio_preprocess: Mel spectrogram ready - n_len=610, n_mel=80, total_size=48800
[CPP] build_whisper: Whisper encoder graph built (projector output, pre-pooling)
[CPP] audition_audio_batch_encode: Audio input data - n_step=610, n_mel=80, total_elements=48800
Desktop (please complete the following information):
| 设备 | Mac air M1, 16GB |
|---|---|
| macOS | 14.8.3 |
| Docker Desktop | 4.58.1 |
| Python | 3.14.2 |
| 模型 | MiniCPM-o-4_5 Q4_K_M |
| 部署模式 | duplex（全双工） |
| Docker 镜像 | 原始镜像 |