mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-02 04:31:25 +00:00

Files

carlushuang 6df5fe2ad8 [CK_TILE]naive attn support FP8 KVCache quant (#1747 )

* quant

* fix bug

* simple smoothquant after softmax

* update kv-quant

* update stride

* fix fp8-pertoken-kvcache

* update int8/fp8 quant support

---------

Co-authored-by: so <a.com>
Co-authored-by: Po Yen Chen <PoYen.Chen@amd.com>

2025-01-03 18:43:07 +08:00

naive_attention.hpp

[CK_TILE]naive attn support FP8 KVCache quant (#1747 )

2025-01-03 18:43:07 +08:00

README.md

[CK_TILE] naive attn (#1708 )

2024-12-12 11:54:03 +08:00

README.md

reference

this folder contains reference implementation of a specific op. Note by including a specific header, you are including the implementation(expecially the gpu implementation) into your source code, and compile that kernel into the fatbin, hence may increase your kernel obj code length. Usually the header starts with reference_ is a cpu reference implementation. The header starts with naive_ contains a gpu implementation with a small launcher.

TODO: move host/reference under this folder