kda-neuron-kernels / build /torch-neuron /nki_kda_decode_batch.py

Commit History

v1.5: exact chunked backward (kda_chunk_step_exact_bwd)
6709b3f

Jim Burtoft commited on

v1.4: forward optimizations (decode key-fold, batched decode, exact prefill, head-interleaved prefill)
5532501

Jim Burtoft commited on