Exact command line arguments for generation
rwzhao opened this issue · comments
When I tried to reproduce the results following the instructions in the README, I failed on the generation after successful training. I ran the following commands, where trained_weights is the directory to save models.
python3 train.py -save=trained_weights
python3 generator.py -data=preprocessed.test.tsv -save=trained_weights/19.vloss-3.649728.lr-0.1
Then I got the error as shown below,
Traceback (most recent call last):
File "generator.py", line 71, in <module>
m.load_state_dict(cpt)
File "/hb/software/apps/python/gnu-3.6.5GPU/lib/python3.6/site-packages/torch/nn/modules/module.py", line 719, in load_state_dict
self.__class__.__name__, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for model:
size mismatch for emb.weight: copying a param of torch.Size([1914, 500]) from checkpoint, where the shape is torch.Size([11738, 500]) in current model.
size mismatch for out.weight: copying a param of torch.Size([1914, 1000]) from checkpoint, where the shape is torch.Size([11738, 1000]) in current model.
size mismatch for out.bias: copying a param of torch.Size([1914]) from checkpoint, where the shape is torch.Size([11738]) in current model.
size mismatch for le.seqenc.lemb.weight: copying a param of torch.Size([6173, 500]) from checkpoint, where the shape is torch.Size([53343, 500]) in current model.
It seems like the vocab size of the testing set is changed. Actually, I think it's the same issue reported here, but I can't get the exact solution from these comments. Could you please provide the exact command line arguments to figure it out? Thanks!
Please pull the repo down with todays changes and confirm if the error persists. Thanks