[Bugfix] Fix qwen2.5-vl overflow issue (#13968)

Signed-off-by: Isotr0py <2037008807@qq.com>

vllm/model_executor/models/utils.py

@@ -641,3 +641,13 @@ def extract_layer_index(layer_name: str) -> int:
     assert len(int_vals) == 1, (f"layer name {layer_name} should"
                                 " only contain one integer")
     return int_vals[0]
+
+
+def cast_overflow_tensors(
+    tensors: torch.Tensor,
+    offset: float = 1000,
+) -> torch.Tensor:
+    if tensors.isinf().any() or tensors.isnan().any():
+        clamp_value = torch.finfo(tensors.dtype).max - offset
+        tensors = torch.clamp(tensors, min=-clamp_value, max=clamp_value)
+    return tensors