aayux / fitfast Goto Github PK

View Code? Open in Web Editor NEW

0.0 0.0 0.0 113 KB

Quickly fine-tune language models for your downstream NLP tasks.

License: Apache License 2.0

Python 100.00%

language-model natural-language-processing transfer-learning transformer-encoder ulmfit

fitfast's People

Contributors

Watchers

fitfast's Issues

learner.sched.plot( ) not showing/saving plots.

Only half an epoch completed on cycle length 2

Loading data/...i/tmp/trn_ids.npy and data/.../tmp/val_ids.npy
Loading pretrained weights...
epoch:   0%|                                                                  | 0/2 [00:00<?, ?it/sepoch      trn_loss   val_loss   accuracy                                                            
    0      3.473275   4.012411   0.335386  
epoch:  50%|█████████████████████████████                             | 1/2 [00:11<00:11, 11.80s/it]

Zero-sized dimensions while batchifying

Traceback (most recent call last):
  File "finetune.py", line 69, in <module>
    main()
  File "finetune.py", line 21, in main
    loader = ff.Loader(train, val, lang='en', bs=64, bptt=60)
  File "/home/guest/user/fitfast/trainer.py", line 70, in __init__
    self.val = LanguageModelIterator(val_ids, bs, bptt)
  File "/home/guest/user/fitfast/text.py", line 19, in __init__
    self.data = self.batchify(data)
  File "/home/guest/user/fitfast/text.py", line 45, in batchify
    return T(data)
  File "/home/guest/user/fitfast/utils/core.py", line 28, in T
    a = torch.LongTensor(a.astype(np.int64))
RuntimeError: the given numpy array has zero-sized dimensions. Zero-sized dimensions are not supported in PyTorch

Recommend Projects