GitLab

Summary:
Changelog:
- `e330f56`: Add code for the "Pay Less Attention with Lightweight and Dynamic Convolutions" paper
- `5e3b98c`: Add scripts for computing tokenized BLEU with compound splitting and sacrebleu
- update READMEs
- misc fixes
Pull Request resolved: https://github.com/pytorch/fairseq/pull/473

Differential Revision: D13819717

Pulled By: myleott

fbshipit-source-id: f2dc12ea89a436b950cafec3593ed1b04af808e9

Summary: Pull Request resolved: https://github.com/pytorch/fairseq/pull/469

Differential Revision: D13802945

Pulled By: myleott

fbshipit-source-id: b6976506a8336b96ee40505c4a7638541cc99c95

Summary:
On a multi-gpu training scenario, the `train.py` script spawns new processes with `torch.multiprocessing.spawn`. Unfortunately those child processes don't inherit the modules imported with `--user-dir`.

This pull request fixes this problem: custom module import in now explicit on every `main()` function.
Pull Request resolved: https://github.com/pytorch/fairseq/pull/449

Differential Revision: D13676922

Pulled By: myleott

fbshipit-source-id: 520358d66155697885b878a37e7d0484bddbc1c6

Summary: Pull Request resolved: https://github.com/pytorch/fairseq/pull/439

Differential Revision: D13608151

Pulled By: myleott

fbshipit-source-id: 198b84995a6329f8329829cc91184d88f1eab947

Summary:
Pull Request resolved: https://github.com/pytorch/translate/pull/283

Pull Request resolved: https://github.com/pytorch/fairseq/pull/428

Differential Revision: D13564190

Pulled By: myleott

fbshipit-source-id: 3b62282d7069c288f5bdd1dd2c120788cee4abb5

Summary: Pull Request resolved: https://github.com/pytorch/fairseq/pull/425

Differential Revision: D13558340

Pulled By: myleott

fbshipit-source-id: dff8c77027e821d8c80bfbd6a6ccce9ca1a44b78

Summary: This is not a guaranteed solution (since processes may still get out of sync if OOM happens after an all_gather/all_reduce has been done) - but should still make multiprocessing training more robust in practice since it seems we usually OOM early enough.

Reviewed By: myleott

Differential Revision: D13086018

fbshipit-source-id: feb1b01c2eb8818797cfdabc0faac8056ba1b4ee

Summary: Pull Request resolved: https://github.com/pytorch/fairseq/pull/374

Differential Revision: D13116074

Pulled By: myleott

fbshipit-source-id: 485724cc5a40e8360d21e4bf9c35821baa0ddc57

Summary:
Manually port fairinternal fairseq-py pull request #385 [1] to fbcode.

Resolve the merge conflict of removing fp16_trainer per offline discussion with Myle. Also updated codes to make generate.py works.

[1] https://github.com/fairinternal/fairseq-py/pull/385/commits/18fa6e154781cf0c4b1596429dba7e753a545069

Reviewed By: liezl200

Differential Revision: D10052908

fbshipit-source-id: c3c378d78dc1e9ac087c815f359e78c0048ff2f5

Summary:
Changelog:
- `90f52a1`: Support loading subsets of the data on each worker with the `--fix-batches-to-gpus` flag. This should fix #217 and #266.
- `6eda0a9`: Update README for replicating the "Scaling Neural Machine Translation" paper
- `b14c7cf`: Fallback to no_c10d backend for pytorch 0.4.1 (fixes #294)
Pull Request resolved: https://github.com/pytorch/fairseq/pull/295

Differential Revision: D10121559

Pulled By: myleott

fbshipit-source-id: 41c84d0ee4cdd113544b5d3aa38ae8b23acc2c27

- no more FP16Trainer, we just have an FP16Optimizer wrapper
- most of the distributed code is moved to a new wrapper class called DistributedFairseqModel, which behaves like DistributedDataParallel and a FairseqModel at the same time
- Trainer now requires an extra dummy_batch argument at initialization, which we do fwd/bwd on when there's an uneven number of batches per worker. We hide the gradients from these dummy batches by multiplying the loss by 0
- Trainer.train_step now takes a list of samples, which will allow cleaner --update-freq

also don't crash if param does not recieve grads

adds -reset-optimizer, --reset-lr-scheduler, and --optimizer-overrides flags

This implements transformer based language model. It already obtains better perplexity on wikitext103 without any tuning. I will also train it on gbw where I also expect to get better ppl

Example training command:

python train.py /private/home/abaevski/data/wiki103 —save-dir /tmp —fp16 —max-epoch 80 —save-interval 1 —arch transformer_lm —task language_modeling —optimizer nag —lr 0.008 —lr-scheduler reduce_lr_on_plateau —lr-shrink 0.6 —dropout 0.2 —criterion adaptive_loss —adaptive-softmax-cutoff 10000,50000,200000 —max-tokens 512 —tokens-per-sample 512 —seed 1 —sample-break-mode none —log-format json —log-interval 50 —save-interval-updates 2500 —keep-interval-updates 25
small transformer got to 31.3 ppl on wiki text 103 (compared to 35 with fconv) while @myleott got a big transformer lm to 27 something ppl on wiki text 103

Two tiny changes to train/eval_lm. For train fix an off by one, while for eval_lm make it work when the task is translation'

A Task defines the data format, stores shared state (e.g., dictionaries) and provides helpers for building the model/criterion and calculating the loss.

Changes:
- Add TranslationTask and LanguageModelingTask. New tasks can be registered with @register_task decorator.
- Add EpochBatchIterator to encapsulate batching and saving/restoring dataloader position
- Remove LEFT_PAD_* constants and make them configurable per task

save best val loss in checkpoint and also print best so far

this way when training continues from an existing checkpoint, we dont immediately override checkpoint_best with a worse loss

This implements convolutional language model from https://arxiv.org/pdf/1612.08083.pdf

There are 3 modes for constructing batches:

- token block: fill each sample with a specified number of tokens without regard for sentence delimiters - this is what was used for training in the paper
- complete: fill each sample with a specified number of tokens but make sure it contains only complete sentences (i.e. if next sentence goes over token block limit, move it to the next sample) - this was used for evaluation in the paper
- eos: one sentence per sample (skip blank lines)

some results:

GCNN-13 - GBW - 37.46
GCNN-14B - GBW - 33.88
GCNN-8 - Wiki103 - 43.76
GCNN-14 - Wiki103 - 35.66

train:

python train.py /private/home/abaevski/data/wiki103 --save-dir /tmp --fp16 --max-epoch 35 --save-interval 1 --save-interval-updates 1000 --keep-interval-updates 25 --arch fconv_lm --optimizer nag --lr 1.0 --lr-scheduler reduce_lr_on_plateau --lr-shrink 0.5 --decoder-embed-dim 280 --decoder-layers '[(850, 6)] * 3 + [(850,1)] + [(850,5)] * 4 + [(850,1)] + [(850,4)] * 3 + [(1024,4)] + [(2048, 4)]' --clip-norm 0.1 --dropout 0.2 --weight-decay 5e-06 --criterion cross_entropy --max-tokens 1024 --max-target-positions 1024 --seed 1 --log-format json --log-interval 500

eval:

python eval_lm.py ~abaevski/data/wiki103 --path '/checkpoint02/abaevski/2018-04-27/lm_wiki.fp16.mxup300000.fconv.adam.lrs=reduce_lr_on_plateau.emb280.layers(850,6)*3+(850,1)+(850,5)*4+(850,1)+(850,4)*3+(1024,1)+(2048,4).lr0.0005.clp0.1.drp0.3.wd0.0.crt=cross_entropy.mxtk2048.smptk256.seed1.ngpu8/checkpoint_last.pt'