[BART] Cannot copy out of meta tensor; no data!

[jiqing-feng]

System Info

Copy-and-paste the text below in your GitHub issue and FILL OUT the two last points.

- `transformers` version: 4.49.0
- Platform: Linux-4.18.0-425.3.1.el8.x86_64-x86_64-with-glibc2.39
- Python version: 3.12.3
- Huggingface_hub version: 0.28.1
- Safetensors version: 0.5.2
- Accelerate version: 1.4.0
- Accelerate config:    - compute_environment: LOCAL_MACHINE
        - distributed_type: MULTI_GPU
        - mixed_precision: bf16
        - use_cpu: False
        - debug: False
        - num_processes: 2
        - machine_rank: 0
        - num_machines: 1
        - gpu_ids: 5,6
        - rdzv_backend: static
        - same_network: True
        - main_training_function: main
        - enable_cpu_affinity: False
        - downcast_bf16: no
        - tpu_use_cluster: False
        - tpu_use_sudo: False
        - tpu_env: []
- DeepSpeed version: not installed
- PyTorch version (GPU?): 2.6.0a0+ecf3bae40a.nv25.01 (True)
- Tensorflow version (GPU?): not installed (NA)
- Flax version (CPU?/GPU?/TPU?): not installed (NA)
- Jax version: not installed
- JaxLib version: not installed
- Using distributed or parallel set-up in script?: <fill in>
- Using GPU in script?: <fill in>
- GPU type: NVIDIA A100 80GB PCIe

Who can help?

@SunMarc @ArthurZucker

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

from transformers import BartForConditionalGeneration
model = BartForConditionalGeneration.from_pretrained("facebook/bart-large-cnn", device_map="cuda:0")

Traceback

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python3.12/dist-packages/transformers/modeling_utils.py", line 262, in _wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/transformers/modeling_utils.py", line 4397, in from_pretrained
    dispatch_model(model, **device_map_kwargs)
  File "/usr/local/lib/python3.12/dist-packages/accelerate/big_modeling.py", line 496, in dispatch_model
    model.to(device)
  File "/usr/local/lib/python3.12/dist-packages/transformers/modeling_utils.py", line 3162, in to
    return super().to(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1344, in to
    return self._apply(convert)
           ^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 904, in _apply
    module._apply(fn)
  File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 904, in _apply
    module._apply(fn)
  File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 931, in _apply
    param_applied = fn(param)
                    ^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1337, in convert
    raise NotImplementedError(
NotImplementedError: Cannot copy out of meta tensor; no data! Please use torch.nn.Module.to_empty() instead of torch.nn.Module.to() when moving module from meta to a different device.

Expected behavior

This is weird because other bart models like distilbart-cnn-12-6 works well.

[Rocketknight1]

I can reproduce this error, and it's quite strange! cc @muellerzr @SunMarc since the error is triggered inside accelerate.

Loading the model on CPU and calling model.to("cuda") works fine for me.

[SunMarc]

I'll have a look. This is probably an issue with weights that are not loaded correctly due to tied weights