Merge branch 'v0.6.2-dev' of ssh://10.6.10.68:10022/dcutoolkit/deeplearing/vllm into v0.6.2-dev

vllm/worker/model_runner.py

@@ -855,11 +855,12 @@ class ModelInputForGPUBuilder(ModelRunnerInputBuilderBase[ModelInputForGPU]):
         tree_attention_masks_list = []
         for inter_data in self.inter_data_list:
             for i in range(len(inter_data.seq_lens)):
-                tree_attn_masks = inter_data.tree_attn_masks[i]
-                if tree_attn_masks is not None:
-                    tree_attention_masks_list.append(tree_attn_masks)
+                if inter_data.tree_attn_masks:
+                    tree_attn_masks = inter_data.tree_attn_masks[i]
+                    if tree_attn_masks is not None:
+                        tree_attention_masks_list.append(tree_attn_masks)
         tree_attention_masks_tensor = None
-        if len(tree_attention_masks_list) > 0:
+        if tree_attention_masks_list:
             tree_attention_masks_tensor = torch.stack(tree_attention_masks_list, dim=0)
             tree_attention_masks_tensor = tree_attention_masks_tensor.contiguous()