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

[ROCm] Enable chunked prefill/paged attention in MLA on ROCm (#14316)

Browse Source 
Signed-off-by: Sage Moore <sage@neuralmagic.com>
This commit is contained in:
Sage Moore
2025-03-12 08:51:20 -07:00
committed by GitHub
parent 4a754fcf15
commit d9f83d6206
2 changed files with 4 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/config.py
View File

@@ -3450,9 +3450,9 @@ class VllmConfig:
self.compilation_config.level = CompilationLevel.NO_COMPILATION
if self.model_config and self.model_config.use_mla and \
not current_platform.is_cuda():
not (current_platform.is_cuda() or current_platform.is_rocm()):
logger.info(
"MLA is enabled on a non-cuda platform; forcing chunked "
"MLA is enabled on a non-GPU platform; forcing chunked "
"prefill and prefix caching to be disabled.")
self.scheduler_config.enable_chunked_prefill = False
self.scheduler_config.chunked_prefill_enabled = False