Pinned
712 contributions in the last year
Contribution activity
January 2021
Created 69 commits in 6 repositories
Created 3 repositories
- sdtblck/DeepSpeedExamples Python
- sdtblck/DALLE-pytorch Python
- sdtblck/DeepSpeed Python
Created a pull request in lucidrains/DALLE-pytorch that received 5 comments
change gumbel softmax dimension
somewhat certain we should be softmaxing over hdim here - pytorch's default is -1 which is the height dimension
+1
−1
•
5
comments
Opened 14 other pull requests in 4 repositories
EleutherAI/gpt-neox
9
merged
- fix everything that i broke
- Pipeline parallelism for enwik8
- implement gradient checkpointing
- Revert GPT2Dataset back to old working state
- Add enron_jsonl and enron_tfr datasets (mostly for testing)
- update tensorflow to 2.4.0
- Fix deprecation warning
- Add improved data downloading class / pipeline
- Update requirements.txt
EleutherAI/gpt-neo
3
merged
EleutherAI/DALLE-mtf
1
merged
microsoft/DeepSpeed
1
open
Reviewed 11 pull requests in 3 repositories
EleutherAI/gpt-neox 7 pull requests
EleutherAI/DALLE-mtf 3 pull requests
EleutherAI/gpt-neo 1 pull request
Created an issue in EleutherAI/gpt-neox that received 6 comments
Implement Pipeline Parallelism
Should be fairly easy as our net is already expressed in terms of layers https://www.deepspeed.ai/tutorials/pipeline/
6
comments
Opened 8 other issues in 2 repositories
EleutherAI/gpt-neox
4
open
3
closed
- Implement Generation / Eval with deepspeed model engine
- Ensure learning rate scheduler is functioning correctly
- Add Deepspeed Transformer Kernel
- Fix tfrecord dataset to load less files into memory
- Write dataset class that tokenizes on the fly
- Openwebtext2 dataset checks for presence of tar.gz file to assess whether to auto-download rather than extracted dataset
- Dataset downloads <number of GPUs> times when running deepspeed train.py

