Update pipeline_qwenimage_edit.py

pipeline_qwenimage_edit.py

@@ -276,37 +276,35 @@ class QwenImageEditPipeline(DiffusionPipeline, QwenImageLoraLoaderMixin):
             )
             texts.append(text)
 
-        # Process inputs - 修复：使用 text_encoder 的实际设备
+        # Process inputs
         model_inputs = self.processor(
             text=texts,
             images=images,
             do_resize=False,   # already resized
             padding=True,
             return_tensors="pt"
-        ).to(device)
+        )
+        
+        # 修复：明确将每个张量移动到正确的设备
+        # 不依赖 .to(device) 的自动传播
+        input_ids = model_inputs.input_ids.to(device)
+        attention_mask = model_inputs.attention_mask.to(device)
+        pixel_values = model_inputs.pixel_values.to(device=device, dtype=dtype)
+        image_grid_thw = model_inputs.image_grid_thw.to(device)
 
         # template = self.prompt_template_encode
         drop_idx = self.prompt_template_encode_start_idx
-        # txt = [template.format(e) for e in prompt]
-
-        # model_inputs = self.processor(
-        #     text=txt,
-        #     images=image,
-        #     padding=True,
-        #     return_tensors="pt",
-        # ).to(device)
 
         outputs = self.text_encoder(
-            input_ids=model_inputs.input_ids,
-            attention_mask=model_inputs.attention_mask,
-            pixel_values=model_inputs.pixel_values,
-            image_grid_thw=model_inputs.image_grid_thw,
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            pixel_values=pixel_values,
+            image_grid_thw=image_grid_thw,
             output_hidden_states=True,
         )
-        # import pdb; pdb.set_trace()
 
         hidden_states = outputs.hidden_states[-1]
-        split_hidden_states = self._extract_masked_hidden(hidden_states, model_inputs.attention_mask)
+        split_hidden_states = self._extract_masked_hidden(hidden_states, attention_mask)
         split_hidden_states = [e[drop_idx:] for e in split_hidden_states]
         attn_mask_list = [torch.ones(e.size(0), dtype=torch.long, device=e.device) for e in split_hidden_states]
         max_seq_len = max([e.size(0) for e in split_hidden_states])