trax resnet50 example doesn't download imagnet dataset, so doesn't work

tensorflow / tensor2tensor

Library of deep learning models and datasets designed to make deep learning more accessible and accelerate ML research.

Apache License 2.0

15.56k stars 3.51k forks source link

trax resnet50 example doesn't download imagnet dataset, so doesn't work #1694

Open jlebar opened 5 years ago

jlebar commented 5 years ago

Description

I'm trying to follow the trax examples in the trax README.md. I ran

python -m tensor2tensor.trax.trainer \
  --config_file=$PWD/trax/configs/resnet50_imagenet_8gb.gin

and this fails with

ValueError: No data files found in /usr/local/google/tmp/trax/image_imagenet-train*

apparently it didn't download the imagenet dataset. The mnist example from the README works (and downloads the dataset).

Full log: https://gist.github.com/9d4268c04465e5ca8bbf0fbd372c6370

Environment information

OS: macos

$ pip3 freeze | grep 'tensor|jax'
tpip3 freeze | grep 'tensor\|jax'
jax==0.1.44
jaxlib==0.1.28
tensorboard==1.14.0
tensorflow-datasets==1.2.0
tensorflow-estimator==1.14.0
tensorflow-gpu==1.14.0
tensorflow-metadata==0.14.0
tensorflow-probability==0.7.0

$ python3 -V
Python 3.6.8rc1

$ git rev-parse HEAD
648184c46abb86239d8fdb317d7921d4bae8800c

jlebar commented 5 years ago

Same for the TransformerDecoder on LM1B example; it also doesn't download the dataset and fails in the same way.

afrozenator commented 5 years ago

This seems related to #1692 fix should be the same? @lukaszkaiser

jlebar commented 5 years ago

This seems related to #1692

Could be, but I don't think so. I worked around #1692 by adding an explicit data dir and it still doesn't download the training data anywhere.

afrozenator commented 5 years ago

Yes, it won't, we'll have to explicitly ask it to f generate_data

On Sun, Sep 8, 2019, 10:38 AM Justin Lebar notifications@github.com wrote:

This seems related to #1692 https://github.com/tensorflow/tensor2tensor/issues/1692

Could be, but I don't think so. I worked around #1692 https://github.com/tensorflow/tensor2tensor/issues/1692 by adding an explicit data dir and it still doesn't download the training data anywhere.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/tensorflow/tensor2tensor/issues/1694?email_source=notifications&email_token=AAIUEFRSJAWHQHXGFJT2G3TQIU2DBA5CNFSM4IUTPC72YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD6FVKMI#issuecomment-529225009, or mute the thread https://github.com/notifications/unsubscribe-auth/AAIUEFQDXC4UO5YDG5F7LZ3QIU2DBANCNFSM4IUTPC7Q .

jlebar commented 5 years ago

I'm happy to do that if it's not too hard, but I'd need a pointer or two. I'm not really sure how all of this is supposed to work. :)