ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-01 20:21:23 +00:00
Code Issues Packages Projects Releases Wiki Activity

[CK_TILE] support group from cmdline (#1295)

Browse Source 
* support cmdline seqlen decode

* silent print

* update readme

* update kernel launch 3d

* update tile partitioner

* fix spill for bf16

* modify based on comment

* modify payload_t

* fix bug for alibi mode

* fix alibi test err

* refactor kernel launch, support select timer

* add missing file

* remove useless code

* add some comments
This commit is contained in:
carlushuang
2024-05-28 11:13:21 +08:00
committed by GitHub
parent 02fa2c298b
commit 5055b3bdcb
16 changed files with 534 additions and 261 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
example/ck_tile/01_fmha/README.md
View File

@@ -34,6 +34,7 @@ args:
if not equal to h, then this is GQA/MQA case
-s seqlen_q. if group-mode, means the average value of seqlen_q (default:3328)
total_seqlen_q = seqlen_q * batch, and seqlen_q per batch may vary
also with "-s=s0,s1,s2..." comma seperated int to set per batch seqlen(group-mode)
-s_k seqlen_k, -1 means equal to s (default:-1)
-d head dim for q, k (default:128)
-d_v head dim for v, -1 means equal to d (default:-1)