flash-attention/csrc/xentropy

This CUDA extension implements optimized cross-entropy loss, adapted from Apex's Xentropy. We make it work for bfloat16 and support in-place backward to save memory.

It has only been tested on A100s.

cd csrc/xentropy && pip install .

As of 2023-09-15, this extension is no longer used in the FlashAttention repo. We've instead switched to a Triton-based implementation. See the CrossEntropyLoss module for more details.