Ross Taylor

We have a leaderboard here https://paperswithcode.com/sota/multi-task-language-understanding-on-mmlu. Feel free to merge this PR if you think this would be useful to link to!

Including SoTA badges for this implementation that highlights the performance of the paper, and allows people to compare to alternative methods.

Currently the user must specify --dev in addition to --cloud to do training on a development instance - an instance that does not close after training. Since the development machine...

File "/home/ubuntu/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages/mantraml/data/Dataset.py", line 288, in extract_file_data file_hashes = self.get_data_dependency_hashes(is_extract_folder=is_extract_folder, is_hash=is_hash) File "/home/ubuntu/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages/mantraml/data/Dataset.py", line 401, in get_data_dependency_hashes self.extract_tar_file(file_path) File "/home/ubuntu/anaconda3/envs/tensorflow_p36/lib/python3.6/site-packages/mantraml/data/Dataset.py", line 349, in extract_tar_file os.rename('%sraw/%s' % (self.data_dir, top_level_dir), '%sraw/%s' % (self.data_dir,...

Errors out with an OverFlow file when the dataset dependencies are large files

Seem to raise an indexing error - something to fix for the next release

Guide on tuning and common problems

Should be able to select/create arrays of latent variables by a name/tag.

There is a mismatch and opening observation is being sliced off.

Should be a method available; possibly a warning presented to the user as well if the model is not invertible after being fitted.