Download Latest Version
v0.19.6 Patch Release source code.zip (218.3 MB)
Get an email when there's a new version of DeepSpeed
| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| README.md | 2026-08-09 | 1.4 kB | |
| v0.19.5 Patch Release source code.tar.gz | 2026-08-09 | 216.9 MB | |
| v0.19.5 Patch Release source code.zip | 2026-08-09 | 218.2 MB | |
| Totals: 3 Items | 435.1 MB | 4 | |
What's Changed
- Update version.txt after 0.19.4 release by @loadams in https://github.com/deepspeedai/DeepSpeed/pull/8219
- Restore the custom DCO workflow by @tohtana in https://github.com/deepspeedai/DeepSpeed/pull/8220
- Update actions versions for workflows. by @loadams in https://github.com/deepspeedai/DeepSpeed/pull/8228
- Do not let one op builder's compatibility probe break importing deepspeed by @adityasingh2400 in https://github.com/deepspeedai/DeepSpeed/pull/8216
- Fix ZeRO++ secondary shard copy for small params by @ZyEng-Art in https://github.com/deepspeedai/DeepSpeed/pull/8210
- fix(zero3): async grad offload + pinned offload buffers by default by @delock in https://github.com/deepspeedai/DeepSpeed/pull/8207
- Fix AutoEP ZeRO-1/2 universal conversion by @tohtana in https://github.com/deepspeedai/DeepSpeed/pull/8198
- Fix compilation error for torch 2.12+ by @hwchen2017 in https://github.com/deepspeedai/DeepSpeed/pull/8238
- Split native host pinning into standalone pin_memory op by @sfc-gh-truwase in https://github.com/deepspeedai/DeepSpeed/pull/8236
- Unmanaged gradient accumulation: ZeRO offload support by @sfc-gh-truwase in https://github.com/deepspeedai/DeepSpeed/pull/8225
New Contributors
- @ZyEng-Art made their first contribution in https://github.com/deepspeedai/DeepSpeed/pull/8210
Full Changelog: https://github.com/deepspeedai/DeepSpeed/compare/v0.19.4...v0.19.5
