accelerate>=1.12.0
einops>=0.8.1
huggingface_hub
librosa>=0.11.0
numpy
openai-whisper
qwen-asr
sentencepiece>=0.2.1
soundfile>=0.13.1
tiktoken>=0.12.0
torch>=2.9.1
torchaudio>=2.9.1
transformers>=4.57.0


# flash-attn>=2.8.0 (optional)  # CUDA-only, speeds up attention
# sageattention>=1.0.0 (optional)  # CUDA-only, experimental attention backend
