[Qwen3VL] fix: hidden_states in place modification error (#41535)

```
  File "transformers/models/qwen3_vl_moe/modeling_qwen3_vl_moe.py", line 941, in forward
    hidden_states = self._deepstack_process(
                    ^^^^^^^^^^^^^^^^^^^^^^^^
  File "transformers/models/qwen3_vl_moe/modeling_qwen3_vl_moe.py", line 960, in _deepstack_process
    hidden_states[visual_pos_masks, :] = local_this
    ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Output 0 of SliceBackward0 is a view and is being modified inplace. This view was created inside a custom Function (or because an input was returned as-is) and the autograd logic to handle view+inplace would override the custom backward associated with the custom Function, leading to incorrect gradients. This behavior is forbidden. You can fix this by cloning the output of the custom Function.
```

Signed-off-by: Hollow Man <[email protected]>

Author: HollowMan6

src/transformers/models/qwen3_omni_moe/modeling_qwen3_omni_moe.py

@@ -1694,7 +1694,8 @@ def _deepstack_process(self, hidden_states, visual_pos_masks, visual_embeds):
         visual_pos_masks = visual_pos_masks[..., 0]
         visual_pos_masks = visual_pos_masks.to(hidden_states.device)
         visual_embeds = visual_embeds.to(hidden_states.device, hidden_states.dtype)
-        local_this = hidden_states[visual_pos_masks, :].clone() + visual_embeds
+        hidden_states = hidden_states.clone()
+        local_this = hidden_states[visual_pos_masks, :] + visual_embeds
         hidden_states[visual_pos_masks, :] = local_this
         return hidden_states
 
@@ -2888,7 +2889,8 @@ def _deepstack_process(
     ):
         visual_pos_masks = visual_pos_masks.to(hidden_states.device)
         visual_embeds = visual_embeds.to(hidden_states.device, hidden_states.dtype)
-        local_this = hidden_states[visual_pos_masks, :].clone() + visual_embeds
+        hidden_states = hidden_states.clone()
+        local_this = hidden_states[visual_pos_masks, :] + visual_embeds
         hidden_states[visual_pos_masks, :] = local_this
         return hidden_states
 

src/transformers/models/qwen3_vl/modeling_qwen3_vl.py

@@ -876,7 +876,8 @@ def _deepstack_process(
     ):
         visual_pos_masks = visual_pos_masks.to(hidden_states.device)
         visual_embeds = visual_embeds.to(hidden_states.device, hidden_states.dtype)
-        local_this = hidden_states[visual_pos_masks, :].clone() + visual_embeds
+        hidden_states = hidden_states.clone()
+        local_this = hidden_states[visual_pos_masks, :] + visual_embeds
         hidden_states[visual_pos_masks, :] = local_this
         return hidden_states
 

src/transformers/models/qwen3_vl/modular_qwen3_vl.py

@@ -746,7 +746,8 @@ def _deepstack_process(
     ):
         visual_pos_masks = visual_pos_masks.to(hidden_states.device)
         visual_embeds = visual_embeds.to(hidden_states.device, hidden_states.dtype)
-        local_this = hidden_states[visual_pos_masks, :].clone() + visual_embeds
+        hidden_states = hidden_states.clone()
+        local_this = hidden_states[visual_pos_masks, :] + visual_embeds
         hidden_states[visual_pos_masks, :] = local_this
         return hidden_states
 

src/transformers/models/qwen3_vl_moe/modeling_qwen3_vl_moe.py

@@ -957,7 +957,8 @@ def _deepstack_process(
     ):
         visual_pos_masks = visual_pos_masks.to(hidden_states.device)
         visual_embeds = visual_embeds.to(hidden_states.device, hidden_states.dtype)
-        local_this = hidden_states[visual_pos_masks, :].clone() + visual_embeds
+        hidden_states = hidden_states.clone()
+        local_this = hidden_states[visual_pos_masks, :] + visual_embeds
         hidden_states[visual_pos_masks, :] = local_this
         return hidden_states