Fusing a 27B ternary LLM's whole decode step into one CUDA kernel

1 points | by Jr23_xd an hour ago

No comments yet