marcelomata/fairseq

Message	Author	SHA1	Date
Some metrics are saved, and better prepared training code for mid-epoch checkpoints	Tolstoyevsky	6247b1d41e	5 years ago
Almost ready to output some training metrics	Tolstoyevsky	c3cf75cfcd	5 years ago
tmp - work on saving all losses	Tolstoyevsky	5442c213d2	5 years ago
Added a script to actually use the hyperparams saved in the yaml file	Tolstoyevsky	6dd747f784	5 years ago
Successfully overfit on a single batch with deep universal transformer	Tolstoyevsky	336edbea59	5 years ago
Checkpoint	Tolstoyevsky	d53fe14e80	5 years ago
Merge branch 'dvc-train-single-batch' into universal-decoder-single-batch	Tolstoyevsky	d1ff4c9605	5 years ago
Universal decoder seems ready (still need to fix minor architecture mismatches in the layers, mainly dropout positions)	Tolstoyevsky	fc04b378e3	5 years ago
Partially successful attempt to overfit a small sample	Tolstoyevsky	78f678088e	5 years ago
Print the reason why training stops	Tolstoyevsky	852843f480	5 years ago
Made training set the same as validation set	Tolstoyevsky	fd742676ea	5 years ago
Merge branch 'dvc' into dvc-train-single-batch	Tolstoyevsky	4cd32ec741	5 years ago
Fixed checkpoint output path in train.sh	Tolstoyevsky	76608d807f	5 years ago
Commit wmt14_en_de_token.dvc	Tolstoyevsky	a223dac19d	5 years ago
Merge branch 'dvc' into dvc-train-single-batch	Tolstoyevsky	a851fa5c00	5 years ago
Fixed reference to resume checkpoint	Tolstoyevsky	38a68f542d	5 years ago
Skipped the unzipping stage completely	Tolstoyevsky	f2db784947	5 years ago
Created a training and validation set that should fit in a single batch, to try to overfit it to validate the model is working	Tolstoyevsky	5daf57765b	5 years ago
Gave better names to the tokenization stage, and moved the prep command to a script	Tolstoyevsky	9737e7b55a	5 years ago
Fix path names in resume-checkpoint.dvc	Tolstoyevsky	cf108ce72b	5 years ago
Removed the unzipped files cache	Tolstoyevsky	189ca2d3b4	5 years ago
Refactored the checkpoint moving stage	Tolstoyevsky	4d4157b01b	5 years ago
Training iteration	Guy	c7c8158854	5 years ago
Added training step stub	Tolstoyevsky	e759a87b56	5 years ago
Ran data prep dvc stage	Guy	d67b3d242d	5 years ago
Change logo size	Dean	7d7f706f0c	5 years ago
Calculated BPE using DVC	Guy	95e6328f05	5 years ago
Added missing commoncrawl dependency	Tolstoyevsky	4d3af97b40	5 years ago
Refactor data preparation a bit	Tolstoyevsky	91ba772189	5 years ago
Universal encoder seems ready	Tolstoyevsky	b5fee3ac3e	5 years ago

Newer Older

marcelomata / fairseq forked from Guy/fairseq

marcelomata
/
fairseq
forked from Guy/fairseq