[Bugfix] Fix Dtypes for Pynccl Wrapper (#33030)

Signed-off-by: Robert Shaw <robshaw@redhat.com> Co-authored-by: Robert Shaw <robshaw@redhat.com>
2026-01-26 15:09:32 -05:00
parent c25dbee40d
commit 43a013c3a2
2 changed files with 8 additions and 2 deletions
--- a/vllm/model_executor/layers/fused_moe/flashinfer_cutlass_prepare_finalize.py
+++ b/vllm/model_executor/layers/fused_moe/flashinfer_cutlass_prepare_finalize.py
@@ -224,6 +224,8 @@ class FlashInferAllGatherMoEPrepareAndFinalize(FlashInferCutlassMoEPrepareAndFin
                a1q_scale = None

        if is_nvfp4 and a1q_scale is not None:
+            if a1q_scale.element_size() == 1:
+                a1q_scale = a1q_scale.view(torch.uint8)
            a1q_scale = nvfp4_block_scale_interleave(a1q_scale)

        return a1q, a1q_scale, None, topk_ids, topk_weights