[Models][Qwen3VL] Optimise _validate_and_reshape_mm_tensor (vllm-project#26426)

lgeiger · web-flow · commit 0426e3c5e149 · 2025-10-09T10:25:48.000Z
Signed-off-by: Lukas Geiger &lt;lukas.geiger94@gmail.com&gt;
diff --git a/vllm/model_executor/models/qwen3_vl.py b/vllm/model_executor/models/qwen3_vl.py
@@ -1304,7 +1304,7 @@ def _validate_and_reshape_mm_tensor(
                     f"Got ndim: {mm_input.ndim} "
                     f"(shape={mm_input.shape})"
                 )
-            return torch.concat(list(mm_input))
+            return mm_input.reshape(-1, mm_input.shape[-1])
         else:
             return torch.concat(mm_input)
 

Original file line number	Diff line number	Diff line change
`@@ -1304,7 +1304,7 @@ def _validate_and_reshape_mm_tensor(`
`1304`	`1304`	`f"Got ndim: {mm_input.ndim} "`
`1305`	`1305`	`f"(shape={mm_input.shape})"`
`1306`	`1306`	`)`
`1307`		`- return torch.concat(list(mm_input))`
	`1307`	`+ return mm_input.reshape(-1, mm_input.shape[-1])`
`1308`	`1308`	`else:`
`1309`	`1309`	`return torch.concat(mm_input)`
`1310`	`1310`