[ROCm] Add env flags to disable dynamic MXFP4 quant and enable AITER tuned GEMMs for Attention Projection Layers (#39987)

Signed-off-by: Hemanth Acharya <heachary@amd.com>
This commit is contained in:
Hemanth Acharya
2026-04-29 16:07:46 -07:00
committed by GitHub
parent c2fb013312
commit 6841f5dc77
5 changed files with 27 additions and 102 deletions
+1
View File
@@ -999,6 +999,7 @@ environment_variables: dict[str, Callable[[], Any]] = {
# use aiter linear op if aiter ops are enabled
# The following list of related ops
# - scaled_mm (per-tensor / rowwise)
# - use aiter tuned gemms for unquantized gemms
"VLLM_ROCM_USE_AITER_LINEAR": lambda: (
os.getenv("VLLM_ROCM_USE_AITER_LINEAR", "True").lower() in ("true", "1")
),