Transformers: Pegasus: OSError: Unable to load weights from pytorch checkpoint file.

[yxyzzz]

Environment info

• transformers version: 3.0.2
• Platform: macOS-10.14.6-x86_64-i386-64bit
• Python version: 3.8.5
• PyTorch version (GPU?): 1.6.0 (False)
• Tensorflow version (GPU?): 2.2.0 (False)
• Using GPU in script?: No
• Using distributed or parallel set-up in script?: No

Who can help

@sshleifer

Information

Model I am using (Bert, XLNet ...): google/pegasus-cnn_dailymail

The problem arises when using:

import torch
from transformers import PegasusForConditionalGeneration, PegasusTokenizer


torch_device = 'cuda' if torch.cuda.is_available() else 'cpu'

model_name = 'google/pegasus-cnn_dailymail'
tokenizer = PegasusTokenizer.from_pretrained(model_name)
model = PegasusForConditionalGeneration.from_pretrained(model_name).to(torch_device)

Traceback:

RuntimeError                              Traceback (most recent call last)
~/projects/transformers/src/transformers/modeling_utils.py in from_pretrained(cls, pretrained_model_name_or_path, *model_args, **kwargs)
    854             try:
--> 855                 state_dict = torch.load(resolved_archive_file, map_location="cpu")
    856             except Exception:

~/anaconda3/envs/abstractive_summarizer/lib/python3.8/site-packages/torch/serialization.py in load(f, map_location, pickle_module, **pickle_load_args)
    584                 return _load(opened_zipfile, map_location, pickle_module, **pickle_load_args)
--> 585         return _legacy_load(opened_file, map_location, pickle_module, **pickle_load_args)
    586 

~/anaconda3/envs/abstractive_summarizer/lib/python3.8/site-packages/torch/serialization.py in _legacy_load(f, map_location, pickle_module, **pickle_load_args)
    771         assert key in deserialized_objects
--> 772         deserialized_objects[key]._set_from_file(f, offset, f_should_read_directly)
    773         if offset is not None:

RuntimeError: unexpected EOF, expected 10498989 more bytes. The file might be corrupted.

During handling of the above exception, another exception occurred:

OSError                                   Traceback (most recent call last)
<ipython-input-1-1ae6eb884edd> in <module>
      7 model_name = 'google/pegasus-cnn_dailymail'
      8 tokenizer = PegasusTokenizer.from_pretrained(model_name)
----> 9 model = PegasusForConditionalGeneration.from_pretrained(model_name).to(torch_device)

~/projects/transformers/src/transformers/modeling_utils.py in from_pretrained(cls, pretrained_model_name_or_path, *model_args, **kwargs)
    855                 state_dict = torch.load(resolved_archive_file, map_location="cpu")
    856             except Exception:
--> 857                 raise OSError(
    858                     "Unable to load weights from pytorch checkpoint file. "
    859                     "If you tried to load a PyTorch model from a TF 2.0 checkpoint, please set from_tf=True. "

OSError: Unable to load weights from pytorch checkpoint file. If you tried to load a PyTorch model from a TF 2.0 checkpoint, please set from_tf=True. 

[sshleifer]

works for me in in torch 1.5.1. and torch 1.6.
Maybe this is a one off s3 failure?
Can anybody else replicate?

from transformers import PegasusForConditionalGeneration
model = PegasusForConditionalGeneration.from_pretrained(model_name)

[yxyzzz]

I set force_download=True and it worked. Thanks!

[55Ankur55]

I set force_download=True and it worked. Thanks!

can you describe in detail how did you solved the problem

[svjan5]

Just upgrading the PyTorch and TensorFlow version solved the problem for me.