ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-04 21:51:28 +00:00
Code Issues Packages Projects Releases Wiki Activity

[CK_TILE] Add paged-kvcache support in group mode fmha fwd splitkv kernels (#1678)

Browse Source 
* Generate group mode paged-attn kernel

* Enable paged-kvcache + group mode support

* Add missing header: fused_moe.hpp

* Add comment to explain kernel arg usage

* Make error message more clear

* Add comment for confusing data member names

* Add more comment for confusing variable names

* Fix typo in option description
This commit is contained in:
Po Yen Chen
2024-11-21 14:53:10 +08:00
committed by GitHub
parent 6916d8cc03
commit fb1ccfa9df
6 changed files with 95 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
example/ck_tile/01_fmha/utils.hpp
View File

@@ -145,7 +145,7 @@ decode_seqlen(mode_enum mode,
std::string k_val,
std::string k_pad_val,
ck_tile::index_t seqlen_k_min = 0,
bool use_kvcache = false,
bool need_append_kvcache = false,
std::optional<unsigned> seed = std::nullopt)
{
#define _S2I_(str_) static_cast<ck_tile::index_t>(std::atoi((str_).c_str()))
@@ -159,7 +159,7 @@ decode_seqlen(mode_enum mode,
const ck_tile::index_t seqlen_k_max = (k < 0 ? q : k);
std::vector<ck_tile::index_t> seqlen_ks(batch, seqlen_k_max);
if(1 < batch && use_kvcache)
if(1 < batch && need_append_kvcache)
{
// to keep the original s_k value, we always use seqlen_k_max in first batch
randints(std::next(seqlen_ks.begin()),