Error in loading the pretrained model

[AnirudhDagar]

Hi @srush Amazing work with ICLR2020. I enjoyed the conference a lot. :)

I was looking into the visualizations and the code for generating the embeddings using the paper abstracts. I am unable to load the pre-trained model you shared here When I try to run the cache_paper_embeddings.py file, I face the following error:

Loading model
Traceback (most recent call last):
  File "cache_paper_embeddings.py", line 31, in <module>
    model, epoch = load_model(None, "scratch/similarity-model.pt")
  File "/Users/gollum/Desktop/Work/iclr.github.io/recommendations/models.py", line 20, in load_model
    model = torch.load(load_file, map_location=lambda storage, loc: storage)
  File "/Users/gollum/anaconda3/lib/python3.7/site-packages/torch/serialization.py", line 529, in load
    return _legacy_load(opened_file, map_location, pickle_module, **pickle_load_args)
  File "/Users/gollum/anaconda3/lib/python3.7/site-packages/torch/serialization.py", line 692, in _legacy_load
    magic_number = pickle_module.load(f, **pickle_load_args)
_pickle.UnpicklingError: invalid load key, 'v'.

This is imo problem with the saved model keys. If possible can you share steps to run the code for training this model?

Thanks!

[srush]

Hi, You will need to use and install `git lfs` to clone the model. You also will need to run it on GPU (like collab). Here are the commands I ran (maybe you can add a README file?) !apt-get install git-lfs !git lfs install !rm -fr iclr.github.io !git clone https://github.com/ICLR/iclr.github.io !pip install -r iclr.github.io/requirements.txt !pip install -r iclr.github.io/recommendations/requirements.txt

…

On Sun, May 3, 2020 at 3:46 AM Anirudh ***@***.***> wrote: Hi @srush <https://github.com/srush> Amazing work with ICLR2020. I enjoyed the conference a lot. :) I was looking into the visualizations and the code for generating the embeddings using the paper abstracts. I am unable to load the pre-trained model you shared here <https://github.com/ICLR/iclr.github.io/blob/master/recommendations/scratch/similarity-model.pt> When I try to run the cache_paper_embeddings.py file, I face the following error: Loading model Traceback (most recent call last): File "cache_paper_embeddings.py", line 31, in <module> model, epoch = load_model(None, "scratch/similarity-model.pt") File "/Users/gollum/Desktop/Work/iclr.github.io/recommendations/models.py", line 20, in load_model model = torch.load(load_file, map_location=lambda storage, loc: storage) File "/Users/gollum/anaconda3/lib/python3.7/site-packages/torch/serialization.py", line 529, in load return _legacy_load(opened_file, map_location, pickle_module, **pickle_load_args) File "/Users/gollum/anaconda3/lib/python3.7/site-packages/torch/serialization.py", line 692, in _legacy_load magic_number = pickle_module.load(f, **pickle_load_args) _pickle.UnpicklingError: invalid load key, 'v'. This is imo problem with the saved model keys. If possible can you share steps to run the code for obtaining these 2D embeddings for each paper? Thanks! — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#74>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAIYKTWXGKQC4GXUSQQVHLRPUOMZANCNFSM4MYAP2WQ> .

[AnirudhDagar]

I did actually use git lfs earlier. But somehow the model was still not unpickling even with a GPU enabled machine.

I repeated the whole process now just to be double sure. It works as expected. Probably something went wrong earlier.

Thanks! I'll make a PR for the readme and mention the steps explicitly.

[srush]

Yeah, the error you gave was what happens when lfs fails.

…

On Sun, May 3, 2020 at 6:04 PM Anirudh ***@***.***> wrote: I did actually use git lfs earlier. But somehow the model was still not unpickling even with a GPU enabled machine. I repeated the whole process now just to be double sure. It works as expected. Probably something went wrong earlier. Thanks! I'll make a PR for the readme and mention the steps explicitly. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#74 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAIYKQ422WI7UBXZDTHFNDRPXS6LANCNFSM4MYAP2WQ> .