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 3, 16:30 UTC
[CI][AMD][BugFix] Prevent triton compiler error when running test_moe_layer with use_ep = True on ROCm (#40857)
Signed-off-by: Randall Smith <Randall.Smith@amd.com>
rasmith committed
May 14, 2026 at 03:44 UTC
768f4a6f269651fd4c727413fd47f6adb28257fe
1 file changed
+1
-1
vllm/model_executor/layers/fused_moe/experts/triton_moe.py
+1
-1
@@ -237,7 +237,7 @@ class TritonExperts(LoRAExpertsMixin, mk.FusedMoEExpertsModular):
237
hidden_states,
238
w1,
239
intermediate_cache1,
240
- a1q_scale,
240
+ a1q_scale if a1q_scale is not None else self.a1_scale,
241
self.w1_scale,
242
None, # topk_weights
243
sorted_token_ids,