[Rocm][Bugfix] Fix dtype not same for gemm_a4w4 op (#33734)

Signed-off-by: charlifu <charlifu@amd.com>
2026-02-06 13:09:59 -06:00
parent 77c09e1130
commit fe5438200b
1 changed files with 6 additions and 1 deletions
--- a/vllm/model_executor/layers/quantization/quark/schemes/quark_ocp_mx.py
+++ b/vllm/model_executor/layers/quantization/quark/schemes/quark_ocp_mx.py
@@ -118,7 +118,12 @@ try:
                )

                gemm_a4w4(
-                    x_q, weight, x_s, weight_scale.view(x_s.dtype), y, bpreshuffle=True
+                    x_q,
+                    weight.view(x_q.dtype),
+                    x_s,
+                    weight_scale.view(x_s.dtype),
+                    y,
+                    bpreshuffle=True,
                )
            return y[:M]
        else: