transformers
accelerate
torchao
flash-linear-attention