jayhshah

[BUG] CUTLASS python emit utility for sm90 GEMM produces code with incorrect syntax/missing header files

3

**Describe the bug** The sm90 gemm code produced by the `cutlass.emit.pytorch` utility has incorrect syntax and missing header files. **Steps/Code to reproduce bug** Install the CUTLASS python interface via `pip...

This PR adds split KV ("Flash decoding") and GQA parallelization improvements for FA3. Some essential parts of the KV cache API are added as well, including the `cache_seqlens` and `cache_batch_idx`...