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, 02:30 UTC
[Bugfix] Remove assert causing hipErrorStreamCaptureUnsupported (#34455)
Signed-off-by: Jaden Mathias <jaden.mathias@amd.com>
Jaden Mathias committed
Feb 18, 2026 at 13:54 UTC
0e60c925cf8c2060c705953d7f89f40c25f7811d
1 file changed
-1
vllm/model_executor/layers/fused_moe/rocm_aiter_fused_moe.py
-1
@@ -381,7 +381,6 @@ class AiterExperts(mk.FusedMoEPermuteExpertsUnpermute):
381
# TODO(rob): rocm_aiter_fused_experts uses self.quant_config's
382
# a_scales for static quantization. Update this to fit better
383
# with the interface once all quant integrations are complete.
384
- assert a2_scale == self.quant_config.a2_scale
384
385
if expert_tokens_meta is not None:
386
num_local_tokens = expert_tokens_meta.expert_num_tokens