A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 12:15 UTC
[ROCm] remove unsupported archs from rocm triton flash-attention supported list (#17536)
Signed-off-by: Hongxia Yang <hongxia.yang@amd.com>
Hongxia Yang committed
May 1, 2025 at 10:54 UTC
28566d73b3c76aa80548a4fb91b4b97cc02b22bc
1 file changed
+1
-1
vllm/attention/ops/triton_flash_attention.py
+1
-1
@@ -650,7 +650,7 @@ def get_general_autotune_configs():
650
651
652
def has_cdna_target():
653
- ROCM_CDNA_TARGETS = ["gfx940", "gfx941", "gfx942", "gfx90a", "gfx908"]
653
+ ROCM_CDNA_TARGETS = ["gfx942", "gfx90a", "gfx908"]
654
return triton.runtime.driver.active.get_current_target(
655
).arch in ROCM_CDNA_TARGETS
656