VictorSanh
3c24e4bef1
Multi-Gpu loss - Cleaning
2018-11-03 18:03:17 -04:00
VictorSanh
1ba5b58c20
fix typo
2018-11-03 17:10:23 -04:00
VictorSanh
5858e8e4dd
Fix both loss and eval metrics -> more coherence on the loss (eval vs train and tf vs pt)
2018-11-03 16:48:24 -04:00
VictorSanh
cd09cd5b40
Fix import on initalization
2018-11-03 15:38:30 -04:00
VictorSanh
a1af5247e1
Add seed in initialization
2018-11-03 14:00:36 -04:00
VictorSanh
8bd6b235b7
typo on tokenization
2018-11-03 10:27:59 -04:00
VictorSanh
2c55568c40
scatter_ and scatter
2018-11-03 10:27:38 -04:00
VictorSanh
a6efe1235f
Merge pull request #1 from huggingface/multi-gpu-support
...
Create DataParallel model if several GPUs
2018-11-03 10:10:34 -04:00
VictorSanh
5f432480c0
Create DataParallel model if several GPUs
2018-11-03 10:10:01 -04:00
VictorSanh
5889765a7c
Update README.md
2018-11-03 09:18:44 -04:00
VictorSanh
8c932e37f9
Update the comparison notebook
2018-11-03 09:08:05 -04:00
VictorSanh
391a4ec2f3
Small typo in trange
...
I seriously don't understand why they defined num_train_epochs as a float in the originial tf code.
I Will change it at the end to avoir merge conflicts for now.
2018-11-03 08:25:15 -04:00
VictorSanh
5676d6f799
Remove BERT pretraining files for now
2018-11-03 08:17:22 -04:00
VictorSanh
72ab10399f
Fix loss
...
Please review @thomwolf but i think this is equivqlent (and it mimics the loss computation of the original loss)
2018-11-02 18:06:21 -04:00
VictorSanh
25d5ca48e0
Fix scatter LopngTensor
2018-11-02 17:57:46 -04:00
VictorSanh
e6a710f684
device
2018-11-02 17:54:22 -04:00
VictorSanh
3ebf1a13c9
Fix loss computation for indexes bigger than max_seq_length.
2018-11-02 17:49:35 -04:00
VictorSanh
38f740a1d5
Fix bug writing predictions in run_squad_pytorch
2018-11-02 11:29:45 -04:00
VictorSanh
ee29871f8d
Debug run_squad_pytorch
2018-11-02 11:07:32 -04:00
VictorSanh
101eabff90
Debug run_squad_pytorch
2018-11-02 10:44:08 -04:00
VictorSanh
bb0a510330
Print for debug run_squad
2018-11-02 10:16:07 -04:00
VictorSanh
844b2f0e6f
Small update Readme
2018-11-02 08:57:15 -04:00
VictorSanh
beb59080b3
Fix size compatibility for model.forward
...
Error was coming from "modeling_pytorch.py", line 484, in forward: start_loss = loss_fct(start_logits, start_positions) --> ValueError: Expected target size (12, 1), got torch.Size([12])
2018-11-02 04:36:42 -04:00
VictorSanh
8cbe7d6af8
FIX errors in loading eval Dataset in run_squad_pytorch
2018-11-02 04:09:53 -04:00
VictorSanh
833c3a7a25
FIX errors in loading Dataset in run_squad_pytorch
2018-11-02 04:00:00 -04:00
VictorSanh
72d69a4ef4
Update README
2018-11-02 03:37:39 -04:00
VictorSanh
62ac7e9a60
Fix small bug in run_squad_pytorch.py
2018-11-02 03:32:35 -04:00
VictorSanh
98b9771dfe
Quick fix metrics evaluation on run_classif_pytorch
2018-11-02 03:02:06 -04:00
VictorSanh
bf65d4dbb7
Begin Updating the README.md
2018-11-02 02:51:07 -04:00
VictorSanh
b54de837c2
Quick fix on eval accuracy
2018-11-02 02:46:17 -04:00
VictorSanh
1d53f9cb72
Fix cuda compability for evaluation
2018-11-02 01:39:15 -04:00
VictorSanh
1d8511f8f2
FIX small bugs in run_classifier_pytorch.py
2018-11-02 01:12:52 -04:00
VictorSanh
936eb4c3ad
FIX small bugs in run_classifier_pytorch.py
2018-11-02 01:11:25 -04:00
VictorSanh
cc228089ef
Small fix to ensure Python3 compativility.
2018-11-02 00:02:36 -04:00
VictorSanh
7af7f8173b
Fix oubli
2018-11-01 15:44:52 -04:00
VictorSanh
d3a8df6b9f
typos in input_fn_builder
2018-11-01 14:17:55 -04:00
VictorSanh
836faed985
wip
2018-11-01 14:17:12 -04:00
VictorSanh
90d360a7a9
WIP
2018-11-01 12:09:23 -04:00
VictorSanh
555b7d66c9
input_fn_builder WIP
2018-11-01 02:10:46 -04:00
VictorSanh
f8e347b557
Convert all DataProcessors, _truncate_seq_pair and convert_examples_to_features
2018-11-01 01:33:01 -04:00
VictorSanh
b1dade34db
Convert flags to argparse in run_classifier_pytorch.py
2018-11-01 01:05:11 -04:00
VictorSanh