torch>=2.4
transformers>=5.0
huggingface_hub>=0.25
safetensors>=0.4

[fast]
flash-linear-attention

[vllm]
vllm>=0.29
