sgugger
Follow
Popular repositories
-
-
-
-
-
Forked from fastai/fastai
The fast.ai deep learning library, lessons, and tutorials
2,604 contributions in the last year
Less
More
Activity overview
Contributed to
huggingface/transformers,
huggingface/accelerate,
huggingface/notebooks
and 5 other
repositories
Contribution activity
June 2021
Created 48 commits in 4 repositories
Created a pull request in huggingface/transformers that received 3 comments
Fix loss reporting with deepspeed
What does this PR do?
In the deepspeed integration inside Trainer, the loss currently reported is the scaled loss, as in scaled by the loss scaling…
+4
−2
•
3
comments
Opened 13 other pull requests in 3 repositories
huggingface/transformers
2
open
7
merged
- AutoTokenizer: infer the class from the tokenizer config if possible
- [WIP] Clean push to hub API
- Use yaml to create metadata
- Temporarily deactivate torchhub test
- Add video links to the documentation
- Add course banner
- Model card defaults
- Don't log anything before logging is setup in examples
- Properly indent block_size
huggingface/accelerate
3
merged
huggingface/datasets
1
merged
Reviewed 65 pull requests in 2 repositories
huggingface/transformers 63 pull requests
- [WIP] Clean push to hub API
- AutoTokenizer: infer the class from the tokenizer config if possible
- [Docs] fixed broken link
- Use yaml to create metadata
- Marian tatoeba conversion update
-
Adding support for
pipeline("automatic-speech-recognition"). - Model card defaults
- [Flax] Add Beam Search
- [WIP] Add CANINE
- Improve detr
- Temporarily deactivate torch-scatter while we wait for new release
- [testing] ensure concurrent pytest workers use a unique port for torch.dist
- Update AutoModel classes in summarization example
- Enable add_prefix_space on run_ner if necessary
- consistent nn. and nn.functional: part 5 docs
-
[style] consistent nn. and nn.functional: part 3
tests -
[style] consistent nn. and nn.functional: part 4
examples - [style] consistent nn. and nn.functional: part2: templates
- Adding TFWav2Vec2Model
- Ray Tune Integration Updates
- [style] consistent nn. and nn.functional
- Fix for the issue of device-id getting hardcoded for token_type_ids during Tracing [WIP]
- [lm examples] Replicate --config_overrides addition to other LM examples
- Use text_column_name variable instead of "text"
-
Have dummy processors have a
from_pretrainedmethod - Some pull request reviews not shown.
huggingface/accelerate 2 pull requests
Created an issue in microsoft/DeepSpeed that received 5 comments
pip install with building ops succeeds without installing any ops.
I'm trying to run deepseed with cpu adam, and after doing:
DS_BUILD_OPS=1 pip install deepspeed
here the output of ds_report:
--------------------…
5
comments