Daniël de Kok fe41e13b45 Unify attention output handling ... - Always return the hidden states. - Create the output tensor inside the `attention` and `paged_attention` functions. This removes the difference between how the output is handled between attention (output parameter) and paged attention (return value). This also removes the assumption that the attention implementation can write to an output tensor (in preparation of FlashInfer).		2024-08-01 13:41:34 +00:00
..
__init__.py	feat: add ruff and resolve issue (#2262)	2024-07-26 10:29:09 -04:00
common.py	[Major Change][Undecided yet] Move to FlashDecoding instead of PagedAttention kernel. (#1940)	2024-07-01 23:28:00 +02:00
cuda.py	Unify attention output handling	2024-08-01 13:41:34 +00:00
flash_attn_triton.py	feat: add ruff and resolve issue (#2262)	2024-07-26 10:29:09 -04:00
ipex.py	Unify attention output handling	2024-08-01 13:41:34 +00:00
rocm.py	Unify attention output handling	2024-08-01 13:41:34 +00:00