facebookresearch / facebookresearch/fairscale
[test] AdaScale: benchmark test
- Dominant language
- Python
- Stars
- 3.4k
- Forks
- 293
- PR merge metrics
- No merged PRs in 30d
Description
It would be nice to have a benchmark test for adascale with cifar that does the full training or a few epochs just to make sure the gain value and training loss is going correctly etc. Also, it can check the overall speed with and without adascale.
Contributor guide
Research direction
Locate the existing AdaScale tests and any CIFAR or training benchmark entry points; no file or test path is named in the issue. Run the current AdaScale training path first, then compare a few-epoch run with and without AdaScale, checking gain values, training loss, and overall speed. Done means the benchmark covers these comparisons with repeatable results.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- machine-learning, performance, testing
- Issue type
- Feature
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 28/100