Qwen Image 2.1 transformer is incompatible with `torch.compile`

[abel1502]

Describe the bug

QwenImage21Rope features a call to image_pad_mask.tolist(). When the transformer in the Qwen Image 2.1 pipeline is torch.compile-ed and then run, it causes the following error:

Unsupported: Tensor.tolist() with non-integer tensor
  Explanation: Dynamo currently does not support tracing `tolist()` on non-integer tensors.
  Hint: Ensure the input tensor to `tolist()` is an integer type (e.g., int8, int16, int32, int64).

  Developer debug context: call_method TensorVariable() to_list

 For more details about this graph break, please visit: https://meta-pytorch.github.io/compile-graph-break-site/gb/gb0109.html

from user code:
   File "[redacted]\.venv\Lib\site-packages\diffusers\models\transformers\transformer_qwenimage21.py", line 925, in forward
    rotary_emb = self.pos_embed(img_shapes[0], image_pad_mask, device=hidden_states.device)
  File "[redacted]\.venv\Lib\site-packages\diffusers\models\transformers\transformer_qwenimage21.py", line 686, in forward
    is_image_token = image_pad_mask.tolist()

Reproduction

import torch
from diffusers import QwenImage21Pipeline

pipe = QwenImage21Pipeline.from_pretrained(
    "Qwen/Qwen-Image-2.1",
    torch_dtype=torch.bfloat16,
).to("cuda")

# Some of this might be unrelated
torch._inductor.config.conv_1x1_as_mm = True
torch._inductor.config.coordinate_descent_tuning = True
torch._inductor.config.epilogue_fusion = False
torch._inductor.config.coordinate_descent_check_all_directions = True
torch._dynamo.config.capture_scalar_outputs = True

pipe.transformer.compile(
    mode="max-autotune", fullgraph=True,
)

image = pipe(
    prompt="The Mona Lisa or whatever.",
    width=2048, height=2048,
    num_inference_steps=40,
    generator=torch.Generator("cuda").manual_seed(42),
).images[0]

System Info

• 🤗 Diffusers version: 0.41.0.dev0
• Platform: Windows-11-10.0.26200-SP0
• Running on Google Colab?: No
• Python version: 3.14.2
• PyTorch version (GPU?): 2.14.0+cu130 (True)
• Huggingface_hub version: 1.32.0
• Transformers version: 5.17.0
• Accelerate version: 1.15.0
• PEFT version: not installed
• Safetensors version: 0.8.0
• xFormers version: not installed
• Accelerator: NVIDIA GeForce RTX 5090, 32607 MiB
• Using GPU in script?: yes
• Using distributed or parallel set-up in script?: no

Who can help?

@naykun @yiyixuxu , based on https://github.com/huggingface/diffusers/blame/main/src/diffusers/models/transformers/transformer_qwenimage21.py