Releases: mila-iqia/milabench
Releases · mila-iqia/milabench
v0.0.3
v0.0.2
What's Changed
- Cancel older CI runs when new version is available by @Delaunay in #78
- Compare by @Delaunay in #77
- Add docker cleanup by @Delaunay in #76
- Add pin command by @satyaog in #71
- Extend the duration of torchbench benchmarks by @breuleux in #79
- Fix pytorch lightning version by @Delaunay in #80
- Bench tags by @Delaunay in #75
- Change nightly to pytorch 2.0 by @Delaunay in #83
- Major rehaul of the code base by @breuleux in #86
- Add ROCm ci config by @Delaunay in #59
- Fix Docker install, race condition, make skip/gpu_poll configurable by @breuleux in #88
- Duplicate docker file for amd by @Delaunay in #91
- Try to cleanup the CI node by @abergeron in #92
- Remove problematic and irrelevant benchmarks, add timeout by @breuleux in #95
- Integrate a new multi-gpu benchmark using accelerate and deepspeed. by @abergeron in #87
- Add Timm benchmark by @breuleux in #89
- Remove lightning by @Delaunay in #99
- Reintroduce StarGAN benchmark by @breuleux in #94
- Reintroduce Super-SloMo benchmark by @breuleux in #98
- Add Whisper benchmark by @breuleux in #93
- Slim down docker images by @abergeron in #100
- Reform the configuration system by @breuleux in #102
- Add dlrm bench by @satyaog in #101
- Expand tags by @Delaunay in #104
- Misc improvements by @breuleux in #106
- Remove multi-stage build for now by @abergeron in #108
- FP32 vs TF32 vs FP16 by @Delaunay in #107
- Multi-node example preview by @abergeron in #96
- Shorten precision benchmarks by @breuleux in #111
Full Changelog: v0.0.1...v0.0.2
v0.0.1
What's Changed
- Remove ROCm Docker by @Delaunay in #62
- Add docker run workflow by @Delaunay in #61
- Use nightly pytorch build by @Delaunay in #64
- Override pytorch in the docker image by @Delaunay in #66
- Create docker.rst by @Delaunay in #67
- Add Sphinx CI to auto deploy doc by @Delaunay in #72
- Standardize dependency management by @satyaog in #49
- Update CI Tags by @Delaunay in #65
- Update numpy minimum version by @Delaunay in #73
- Use miniconda py3.9 instead of latest by @Delaunay in #74
Full Changelog: v0.0.0...v0.0.1
v0.0.0 H100 Support
Use nightly pytorch build (#64) * Use nightly pytorch build * Force install specific pytorch version * Fix PL bolts --------- Co-authored-by: Pierre Delaunay <[email protected]>