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 8, 12:30 UTC
[Bugfix] Fix benchmark_moe.py inplace assertion with torch >= 2.9 (#34149)
Signed-off-by: Matthias Gehre <matthias.gehre@amd.com>
Matthias Gehre committed
Feb 11, 2026 at 04:58 UTC
7a048ee65f0b8da2c2493ef76cbee89cf612baa6
1 file changed
+3
-2
benchmarks/kernels/benchmark_moe.py
+3
-2
@@ -226,9 +226,10 @@ def benchmark_config(
226
x, input_gating, topk, renormalize=not use_deep_gemm
227
)
228
229
+ inplace = not disable_inplace()
230
if use_deep_gemm:
231
return deep_gemm_experts(
231
- x, w1, w2, topk_weights, topk_ids, inplace=True
232
+ x, w1, w2, topk_weights, topk_ids, inplace=inplace
233
)
234
return fused_experts(
235
x,
@@ -236,7 +237,7 @@ def benchmark_config(
237
w2,
238
topk_weights,
239
topk_ids,
239
- inplace=True,
240
+ inplace=inplace,
241
quant_config=quant_config,
242
)
243