numpy soundfile torch torchaudio transformers>=4.57.0,<4.58.0 accelerate huggingface_hub diffusers einops loguru tqdm numba>=0.63.1 vector-quantize-pytorch PyYAML modelscope filelock>=3.13.0 peft>=0.11.0 gradio>=4.0.0 pandas bitsandbytes fastapi uvicorn python-multipart openai python-dotenv