services: qwen3-tts: image: qwen3-tts-custom:${QWEN_TTS_VERSION:-1.2.1} container_name: qwen3-tts restart: unless-stopped gpus: all environment: # HF cache 仍然保留(Tokenizer 可能还会用到) HF_HOME: /cache/hf HUGGINGFACE_HUB_CACHE: /cache/hf/hub TRANSFORMERS_CACHE: /cache/hf/transformers volumes: # HF 缓存(可选,但推荐) - ./cache:/cache # ✅ 你的本地模型目录,挂载进容器 - ./qwen3-tts-customer:/models/qwen3-tts:ro ports: - "8000:8000" command: > bash -lc " qwen-tts-demo /models/qwen3-tts --ip 0.0.0.0 --port 8000 --no-flash-attn "