Summary:
# Before submitting
- [ ] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [ ] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [ ] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
## What does this PR do?
Fixes # (issue).
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2678
Reviewed By: Mortimerp9
Differential Revision: D32653381
Pulled By: dianaml0
fbshipit-source-id: 2810d14867cd7d64f4d340740e2b590b82de47fe
Summary:
# Before submitting
- [ ] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [ ] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [ ] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
## What does this PR do?
Adds circleCI config to `main` similar to the one added in https://github.com/fairinternal/fairseq-py/issues/2455 by vedanuj
Splitting out changes in https://github.com/fairinternal/fairseq-py/issues/2570 to be cleaner.
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2677
Reviewed By: Mortimerp9
Differential Revision: D32653411
Pulled By: dianaml0
fbshipit-source-id: 24950ecebfb569b552675e65bee1451b8070f8ee
Summary:
# Before submitting
Separating https://github.com/fairinternal/fairseq-py/issues/2212 into separate PR's to be cleaner.
- [ ] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [ ] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [ ] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
## What does this PR do?
Fixes # (issue).
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2676
Reviewed By: alexeib
Differential Revision: D32653505
Pulled By: dianaml0
fbshipit-source-id: fd338289d23b340ef56b0188f9c73b37c367d6ca
Summary:
New version of fairscale breaks with python3.6. Moving to higher versions of python.
# Before submitting
- [ ] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [ ] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [ ] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
## What does this PR do?
Fixes # (issue).
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2679
Reviewed By: alexeib
Differential Revision: D32653241
Pulled By: dianaml0
fbshipit-source-id: bb81023b967f7d585fe9fb808f65cfd60cf51490
Summary:
Fix a typo
# Before submitting
- [ ] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [ ] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [ ] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
## What does this PR do?
Fixes # (issue).
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2664
Reviewed By: alexeib
Differential Revision: D32577174
Pulled By: arbabu123
fbshipit-source-id: 9b65619843fd9684be319024ccc9e6b0681db7de
Summary:
There are three options for the ontology:
* `--use-ontology-for-training`
* `--use-ontology-for-validation`
* `--use-ontology-for-balancing`
The first two must always be set together.
In the past, I observed that it's best not to use ontology for data balancing even if we use ontology for training and validation. But now I no longer observe this.
Therefore, I'm merging all these three options into one (`--use-ontology`).
In addition, I'm also moving the logic of avoiding loading teacher models out of `checkpoint_utils.py`. If you want to load a student model without loading its teachers (e.g. for prediction only), specify `arg_overrides={"ignore_teachers": True}` when calling `load_model_ensemble`.
Reviewed By: xiaoxiao26
Differential Revision: D32518830
fbshipit-source-id: 103c6458f7927ec5ca7470109c8f956c00f514a2
Summary:
# Before submitting
- [ ] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [ ] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [ ] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
## What does this PR do?
Fixes # (issue).
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2658
Reviewed By: ngoyal2707
Differential Revision: D32520446
Pulled By: arbabu123
fbshipit-source-id: a4cbc12624c9c8c1b5bc3d64eb47c2fdec01eb87
Summary:
# Before submitting
- [ ] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [ ] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [ ] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
## What does this PR do?
Fixes # (issue).
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2652
Reviewed By: michaelauli
Differential Revision: D32499609
Pulled By: arbabu123
fbshipit-source-id: 1d91f0c5dda8f4e0d0a8d3c8f4854188e91500c3
Summary:
# Before submitting
- [ ] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [ ] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [ ] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
## What does this PR do?
Fixes # (issue).
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2545
Reviewed By: alexeib
Differential Revision: D32150553
Pulled By: arbabu123
fbshipit-source-id: 1acff92dd69db128bc97ee36ca55b44dd59e08f5
Summary:
- Allows for faster training on multiple GPUs when batches are based on
sorted input sequences.
- Instead of shuffling batches randomly followed by distribution on
workers, we group the batches in sets of total workers and then shuffle
the groups. When the batches are sorted by length this ensures that each
worker receives similar length inputs.
# Before submitting
- [N] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [Y] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [ ] Did you make sure to update the docs?
- [N] Did you write any new necessary tests?
## What does this PR do?
Adds option "grouped_shuffling" to the dataclass to allow batches to be first grouped in set of total workers followed by shuffling of the groups. This reduces the sequence length discrepancy among the workers when the batches were created from inputs sorted by sequence lengths.
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2391
Reviewed By: arbabu123
Differential Revision: D31352971
Pulled By: alexeib
fbshipit-source-id: c045bedecb03339c8eb46e7e8c9804a53b35615b
Summary:
The logits were not correctly masked leading to significant WER
degradation for batched inference with padded inputs. This resolves it.
However there is minor degradation still from somewhere else.
# Before submitting
- [ ] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [x] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [x] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
## What does this PR do?
Fixes logits padding mask for wav2vec asr model.
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2469
Reviewed By: arbabu123
Differential Revision: D32469676
Pulled By: alexeib
fbshipit-source-id: 154bb6523f9300e4b6c1cf2631424fd22f582bc5
Summary:
## What does this PR do?
- Update robust wav2vec2 fine-tuned checkpoints to include dictionary and w2v_args.
- Update README to point to the updated checkpoints
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2636
Reviewed By: alexeib
Differential Revision: D32414361
Pulled By: wnhsu
fbshipit-source-id: e4bf41e78593cde8dc08a1bfacdd3ec5005e54bc
Summary:
this fixes a crash when using num_workers > 0 and num processes > 1 (e.g. by launching on slurm with tasks-per-node > 1 or using torch.distributed/torchrun with num gpus > 1). the root cause is some kind of bug in torch.split() that holds on to memory. when data loaders recycle, the program crashes when trying to release already released memory
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2623
Reviewed By: arbabu123
Differential Revision: D32342431
Pulled By: alexeib
fbshipit-source-id: 278c7e0b09cfd770f375865d472bc52fb5835bd2
Summary:
- [ ] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [x] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [x] Did you make sure to update the docs?
- [x] Did you write any new necessary tests?
## What does this PR do?
SlowMo is being moved to [Fairscale](https://fairscale.readthedocs.io/en/latest/). This commit updates the implementation of SlowMo to the Fairscale version. It also adds tests for SlowMo.
Note: This PR is currently for review. It will be merged at a later date once SlowMo has been updated to Fairscale. SlowMo is being merged to Fairscale as part of [a PR](https://github.com/facebookresearch/fairscale/pull/378). So, once that PR is merged to Fairscale, this PR on Fairseq will be ready for merge
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/pytorch/fairseq/pull/3996
Reviewed By: dianaml0
Differential Revision: D32280163
Pulled By: vtantia
fbshipit-source-id: 70c97b04a7cdc90ada7099375c2a31b0c978ba70
Summary:
## What does this PR do?
Updates the bibtex to cite the mtedx dataset in the speech-to-text example, now that mtedx has been published (Interspeech 2021)
Pull Request resolved: https://github.com/pytorch/fairseq/pull/3984
Reviewed By: yuntang
Differential Revision: D32102153
Pulled By: kahne
fbshipit-source-id: 1eea532e9032c116da447a4da30669305559a787
Summary:
1. Add links to corresponding code_dict files, required by the TTS
checkpoints
2. synthesize_audio_from_units.py has code_dict path configurable
3. Removed links to 5000-unit models, as those will not be supported.
# Before submitting
- [x] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
Fixes https://github.com/pytorch/fairseq/issues/3970 and https://github.com/pytorch/fairseq/issues/3870
- [x] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [ ] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
Manually verified CPC50 & 100, HUBERT 50-200
## What does this PR do?
Fixes issues 3970 and and 3870 in the public repo
1. Add links to corresponding code_dict files, required by the TTS checkpoints
2. synthesize_audio_from_units.py has code_dict path configurable
3. Removed links to 500-unit models, as those will not be supported.
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2519
Reviewed By: wnhsu
Differential Revision: D32019362
Pulled By: eugene-kharitonov
fbshipit-source-id: a2a1db99898e4d025c5a1e439bd63264406aad0a
Summary:
## What does this PR do?
Set `model.num_updates` in `load_model_ensemble_and_task()` before loading `state_dict`, like what's done in `fairseq/trainer.py`, because a model's `state_dict` may depend on `num_update`.
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2491
Reviewed By: xuqiantong
Differential Revision: D31863368
Pulled By: wnhsu
fbshipit-source-id: c70051f898819cc43b02c9f5765429e9f194aed5
Summary:
CPLTaskImpl provides implementation to augment existing tasks to take additional input of ema_model in its train_step and valid_step for continous pseudo-labeling (CPL) during training. It passes this ema_model to the criterion.
See Kaizen semi-supervised training paper for more details https://arxiv.org/abs/2106.07759.
This implementation also supports using CPLDataset which enables using unsupervised data only for `cpl_finetune_epoch > epochs >= cpl_start_epoch`. CPLDataset is like MultiCorpusDataset but ignores the unsupervised datasets while sampling.
Another addition in this diff is to skip dataset in MultiCorpusDataset if the sampling probability is 0.
Reviewed By: cruvadom
Differential Revision: D30701536
fbshipit-source-id: 1d840eacfd538ed7aed3baaefc8b254390642b45
Summary:
Currently the speech recognition fails for fp16 option as the input is
not converted to half. This fixes this issue.
# Before submitting
- [ ] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [x] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [x] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
## What does this PR do?
Fixes fp16 inference for speech recognition example.
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2438
Reviewed By: apoorv2904, xuqiantong
Differential Revision: D31594529
Pulled By: alexeib
fbshipit-source-id: 75d1fe3f00b2e86381f28f79daea2cd516233e22
Summary:
hasattr should not default to True if field doesn't exist
# Before submitting
- [ ] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [ ] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [ ] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
## What does this PR do?
Fixes # (issue).
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/pytorch/fairseq/pull/3953
Reviewed By: myleott, Lynx1820
Differential Revision: D31591929
Pulled By: dianaml0
fbshipit-source-id: f78397d2b06867c61fc04a6bd2aa9579265e2d21
Summary:
# Before submitting
- [x] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [x] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [x] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
## What does this PR do?
Fixes argument for `lr_scheduler.total_num_update`; missing import of `dsprocessor` for COIN; `vmasks` on demo inference; update README.md of fairseq for examples/MMPT.
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2428
Reviewed By: berniebear
Differential Revision: D31528947
Pulled By: howardhsu
fbshipit-source-id: 1fecf34bdab82cbf6001e3905a532e4e6eb38e01
Summary:
# Before submitting
- [x] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [x] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [x] Did you make sure to update the docs?
- [] Did you write any new necessary tests?
## What does this PR do?
Fixes
https://github.com/pytorch/fairseq/issues/3930https://github.com/pytorch/fairseq/issues/3795
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2424
Reviewed By: Abdel-rahmanMohamed
Differential Revision: D31506758
Pulled By: wnhsu
fbshipit-source-id: 895c0b034637f081dccb9905f38aced8a18ce5ff
Summary:
Loading xlmr doesn't work because trying to pull from master branch, which has been changed to main.
Pull Request resolved: https://github.com/pytorch/fairseq/pull/3922
Reviewed By: erichan1
Differential Revision: D31415771
Pulled By: dianaml0
fbshipit-source-id: a4b1b533b5596a4d3556c97c2f15847e444aeea9
Summary:
# Before submitting
- [ ] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [ ] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [ ] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
## What does this PR do?
Fixes # (issue).
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2415
Reviewed By: alexeib
Differential Revision: D31455210
Pulled By: xuqiantong
fbshipit-source-id: d3b45a3b60800f0fc5da0efeddbd8213dee4878f
Summary:
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2200
The expected alignment for p-choose is the performance bottleneck that needs to be optimized. The solution is to implement a custom operator to reduce the kernel launch overhead, and optimize the implementations of some operations.
Some key optimizations:
* Use a contiguous alpha array to avoid array concatenation. The original version create an array for each slice of alpha and concat them in the end.
* Implement cumprod using prod operation directly. It used log-cumsum-exp operations before.
* Implement cumprod using cuda CUB library which is more efficient than scan operation in pytorch.
Reviewed By: cndn
Differential Revision: D30033767
fbshipit-source-id: 853c1c2d366838d6bcfa0863999f217a394e46a7
Summary:
# Before submitting
- [ ] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [ ] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [ ] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
## What does this PR do?
Fixes # (issue).
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2408
Reviewed By: alexeib
Differential Revision: D31422844
Pulled By: xuqiantong
fbshipit-source-id: 348050118621de922c08632ca7f31dedba109da1
Summary:
# Before submitting
- [ ] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [ ] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [ ] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
## What does this PR do?
zero-shot model release
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2407
Reviewed By: alexeib
Differential Revision: D31417241
Pulled By: xuqiantong
fbshipit-source-id: 576644694638d3b2606f1751b74feb0531b50eb7
Summary:
# Before submitting
- [x] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [x] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/main/CONTRIBUTING.md)?
- [x] Did you make sure to update the docs?
- [x] Did you write any new necessary tests?
## What does this PR do?
Release the code and model for two of our papers at FAIR:
1. VideoCLIP: Contrastive Pre-training for Zero-shot Video-Text Understanding (Xu et. al., EMNLP 2021)
2. VLM: Task-agnostic Video-Language Model Pre-training for Video Understanding (Xu et. al., ACL Findings 2021)
## PR review
dianaml0 (Diana Liskovich, referred by Myle Ott)
## Did you have fun?
Yes! {emoji:1f44d}
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2373
Reviewed By: dianaml0
Differential Revision: D31278832
Pulled By: berniebear
fbshipit-source-id: b6a0fad4caf44b062be0c46c12842b26792b35a3
Summary:
There are mismatches for the code in speech_text_joint_to_text example and the code in the latest fairseq codebase
1. import task class twice
2. newly added TransformerEncoderLayerBase is equal to TransformerEncoderLayer
3. Wav2VecEncoder API change (wav2vec2_asr.py)
Reviewed By: kahne
Differential Revision: D31299458
fbshipit-source-id: 6eb64e2692ca3c2729248d55ccefe74283fe4ef0
Summary:
We use omegaconf.DictConfig objects in non-strict mode, so hasattr behaves weirdly:
```
>>> import omegaconf
>>> omegaconf.__version__
'2.0.6'
>>> x = omegaconf.DictConfig({"a": 1})
>>> hasattr(x, "foo")
True
```
This violates some assumptions in various parts of the code. For example, previously this command was incorrectly missing the final layer norm due to upgrade logic that relied on `hasattr`, but is fixed after this diff:
```
CUDA_VISIBLE_DEVICES=0 python train.py --task dummy_lm --arch transformer_lm_gpt3_small --optimizer adam --lr 0.0001 --max-sentences 8 --log-format json --log-interval 1
```
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2347
Reviewed By: alexeib
Differential Revision: D31170584
Pulled By: myleott
fbshipit-source-id: bd767b7497794314f58f0f8073cdd4332b214006
Summary: Update reference from master to main elsewhere in fbcode
Reviewed By: alexeib
Differential Revision: D30938472
fbshipit-source-id: 243b98550207f241c9d3265bf3d4060350aaf0a8
Summary:
…verride the defaults
# Before submitting
- [x] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [x] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/master/CONTRIBUTING.md)?
- [ ] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
## What does this PR do?
Fixes https://github.com/pytorch/fairseq/issues/3761.
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/pytorch/fairseq/pull/3773
Reviewed By: yuntang
Differential Revision: D30310383
Pulled By: kahne
fbshipit-source-id: cbfcbc032dbf53490a25ffdebe57f65c42d52e71
Summary:
# Before submitting
- [ ] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [ ] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/master/CONTRIBUTING.md)?
- [ ] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
## What does this PR do?
Fixes # (issue).
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/2297
Reviewed By: alexeib
Differential Revision: D30906090
Pulled By: dianaml0
fbshipit-source-id: 941d30db7f766c9077a1b5bb2a04680f57e2e070
Summary:
# Before submitting
- [ ] Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
- [ ] Did you read the [contributor guideline](https://github.com/pytorch/fairseq/blob/master/CONTRIBUTING.md)?
- [ ] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
## What does this PR do?
Fixes # (issue).
## PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
## Did you have fun?
Make sure you had fun coding �
Pull Request resolved: https://github.com/pytorch/fairseq/pull/3879
Reviewed By: myleott
Differential Revision: D30969142
Pulled By: dianaml0
fbshipit-source-id: 902154c03fd68ae6645d3e0ac07b7d729dfc7934