Skip to content

Pull requests: pytorch-tpu/fairseq

New pull request New
Clear current search query, filters, and sorts
0 Open 39 Closed
0 Open 39 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

Replacy numpy.float with python's float to fix test pt-nightly-fs-transformer-func-v3-32-1vm
#39 by vanbasten23 was merged Jan 25, 2023 Loading…
2
Modify train.py to run on PjRT
#38 by darisoy was closed Aug 25, 2022 Loading…
Replace offending inplace operation.
#37 by taylanbil was merged Jul 24, 2020 Loading…
Replace usage of deprecated function.
#36 by taylanbil was merged Jun 15, 2020 Loading…
Use the device, not the str representation when instantiating Paralle…
#35 by taylanbil was merged May 18, 2020 Loading…
Write TimeMeters and StopWatchMeters to tensorboard.
#34 by taylanbil was merged Apr 3, 2020 Loading…
1
Mid-training validation and checkpointing.
#33 by taylanbil was merged Apr 3, 2020 Loading…
2
Reduce gradients before clipping.
#32 by taylanbil was merged Mar 31, 2020 Loading…
6
Resolve SEGFAULT with workaround when loading ckpt
#31 by jysohn23 was closed Mar 25, 2020 Loading…
2
Add checkpointing to GCS from global master, clean up checkpointing.
#30 by taylanbil was merged Mar 24, 2020 Loading…
2
Change .view to .reshape in order to fix generate.py
#29 by taylanbil was merged Mar 20, 2020 Loading…
Fix tensorboard logging, and add tboard logging consumable by the OSS metrics handler
#28 by taylanbil was merged Feb 26, 2020 Loading…
4
Make roberta pretraining task work on tpus
#27 by taylanbil was merged Feb 19, 2020 Loading…
8
Enabling tensorboard logging for progress bars
#26 by taylanbil was merged Dec 26, 2019 Loading…
19
Assert using the loss values instead of reset training meters.
#25 by taylanbil was merged Dec 5, 2019 Loading…
1
Sending optimizer and lr_scheduler states to device before loading in
#24 by taylanbil was merged Dec 3, 2019 Loading…
Resetting trainer's meters in tpu training as in the gpu case.
#23 by taylanbil was merged Nov 22, 2019 Loading…
Added a barrier after every gradient accumulation step.
#22 by taylanbil was merged Nov 21, 2019 Loading…
4
Replace log_step with progress_bar printing
#21 by taylanbil was merged Nov 20, 2019 Loading…
3
Remove upgrade_state_dict_named
#20 by taylanbil was merged Nov 19, 2019 Loading…
Rebasing tpu branch on a more recent fairseq upstream commit
#19 by taylanbil was merged Nov 19, 2019 Loading…
47
Rebasing tpu branch on a more recent fairseq upstream commit
#18 by taylanbil was closed Nov 16, 2019 Loading…
1
Enable usage of batch_by_size for other code paths than just train
#17 by taylanbil was merged Nov 7, 2019 Loading…
7
Make 1 checkpoint directory per host.
#16 by taylanbil was merged Nov 6, 2019 Loading…
2
Replace usage of unsqueeze with transpose + broadcasting in multihead attention
#15 by taylanbil was merged Oct 30, 2019 Loading…
1
ProTip! Adding no:label will show everything without a label.