biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[BugFix][V1][ROCm] Triton MLA uses V0 backend on V1 engine (#19067)

Browse Source 
Signed-off-by: Tianyuan Wu <Tianyuan.Wu@amd.com>
This commit is contained in:
TY-AMD
2025-07-01 16:12:19 +08:00
committed by GitHub
parent b1c1fe35a5
commit 96453cfa83
5 changed files with 78 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/platforms/rocm.py
View File

@@ -186,8 +186,14 @@ class RocmPlatform(Platform):
if selected_backend == _Backend.TRITON_MLA:
if block_size != 1:
logger.info("Using Triton MLA backend.")
return "vllm.attention.backends.triton_mla.TritonMLABackend" # noqa: E501
if use_v1:
logger.info_once(
"Using Triton MLA backend on V1 engine.")
return ("vllm.v1.attention.backends.mla."
"triton_mla.TritonMLABackend")
else:
logger.info("Using Triton MLA backend.")
return "vllm.attention.backends.triton_mla.TritonMLABackend" # noqa: E501
else:
raise ValueError(
f" The selected backend, {selected_backend.name},"