GitHub / microsoft/DeepSpeed / commits
DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.
| SHA | Message | Author | Date | Stats |
|---|---|---|---|---|
| 4370aa50 |
feat(zero): enable torch.func transforms on engine for ZeRO 0/1/2 (#8026)
Co-authored-by: Masahiro Tanaka <8****a@u****m> Signed-off-by: Sung Hyun Cho <h****7@g****m> |
Sung Hyun Cho <h****7@g****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| 60b242af |
Add engine.coalesce_grad_reduction() for ZeRO 1/2/3 multi-backward (#7992)
Co-authored-by: Masahiro Tanaka <8****a@u****m> Signed-off-by: Sung Hyun Cho <h****7@g****m> |
Sung Hyun Cho <h****7@g****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| a0b98d34 |
Bump version (#8030)
Co-authored-by: Masahiro Tanaka <8****a@u****m> |
Olatunji Ruwase <t****e@s****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| 77a249f1 |
Keep required CI checks visible for ignored paths (#8019)
Co-authored-by: Olatunji Ruwase <t****e@s****m> Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| 4570c508 |
Fix full CI test isolation for ZeRO chmod and NVMe quantization tests (#8008)
Co-authored-by: Olatunji Ruwase <t****e@s****m> Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| b7aef4dc |
Fix PR-target workflow concurrency groups (#8017)
Co-authored-by: Olatunji Ruwase <t****e@s****m> Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| 2c8a007b |
Run AutoSP compile tests sequentially (#8020)
Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| 01e3a28b |
fix(fp16): filter requires_grad in FP16 optimizer flat buffer init (#8029)
Signed-off-by: Avi Cooper <a****7@g****m> |
avicooper1 <a****7@g****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| 510ebe58 |
Fix DeepCompile AOT kwargs patching for PyTorch >= v2.11 (#8024)
Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| 45429221 |
Support flash-attn 2.7.0 in FPDT attention (#8022)
Co-authored-by: binchengxiong <b****g@a****m>, Claude Opus 4.7 <n****y@a****m>, Masahiro Tanaka <8****a@u****m> Signed-off-by: binchengxiong <b****g@a****m> |
bincheng.xiong <1****5@q****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| ae576f83 |
fix(zero): enable vmap on LinearFunctionForZeroStage3 (#8023)
Co-authored-by: Masahiro Tanaka <8****a@u****m> Signed-off-by: Sung Hyun Cho <h****7@g****m> |
Sung Hyun Cho <h****7@g****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| 43166463 |
Sort and dedupe -gencode flags emitted by op_builder.builder (#8021)
Signed-off-by: Aditya Singh <a****0@g****m> |
Aditya Singh <6****0@u****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| 67b9e212 |
fix: add setup_context for torch.func compatibility (#7916)
Co-authored-by: zhangj1an <j****s@g****m>, Zhang Jian <z****n@u****u>, Masahiro Tanaka <m****a@a****m> Signed-off-by: Sung Hyun Cho <h****7@g****m>, Zhang <j****s@g****m>, Masahiro Tanaka <m****a@a****m>, Zhang Jian <j****s@g****m> |
Sung Hyun Cho <h****7@g****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| bd71eed6 |
fix gemma4 num attention head bugs (from #7975) (#7990)
Signed-off-by: Guokai Ma <g****a@i****m> |
Ma, Guokai <g****a@g****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| de473091 |
[Blog] Muon Optimizer Support in DeepSpeed (#7962)
Signed-off-by: Ma, Guokai <g****a@i****m>, Ma, Guokai <g****a@g****m>, Guokai Ma <g****a@i****m> |
Ma, Guokai <g****a@g****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| 26ee3856 |
[fix] fix test_zf.py hang bug (#8012)
Co-authored-by: binchengxiong <b****g@a****m>, Masahiro Tanaka <8****a@u****m> Signed-off-by: binchengxiong <b****g@a****m> |
bincheng.xiong <1****5@q****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| 3c337b54 |
Support bf16 optimizer states with CPU offload (#8010)
Co-authored-by: Claude Opus 4.7 <n****y@a****m>, Masahiro Tanaka <8****a@u****m> Signed-off-by: Lucas Pirola <l****s@p****u> |
lucaspirola <l****s@p****u>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| 9f95452e |
Make GitHub Actions job names unique (#8014)
Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| d6d4b3d0 |
Run FastFileWriter fd-close test outside pytest-forked (#8015)
Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| b387166f |
Add configurable torch-latest dependency versions (#8016)
Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| ae075141 |
Remove stale step() docstring from DeepSpeedCPUAdam (#8011)
Co-authored-by: Claude Opus 4.7 <n****y@a****m> Signed-off-by: Lucas Pirola <l****s@p****u> |
lucaspirola <l****s@p****u>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| d7a3972f |
Fix ZeRO-3 forward crash on modules with plain dict _parameters (#8009)
Co-authored-by: Masahiro Tanaka <8****a@u****m> Signed-off-by: Sung Hyun Cho <h****7@g****m> |
Sung Hyun Cho <h****7@g****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| 8cdf8651 |
fix: use subprocess instead of os.system in data_analyzer.py (#7994)
Co-authored-by: Masahiro Tanaka <8****a@u****m> Signed-off-by: orbisai0security <m****y@g****m> |
OrbisAI Security <m****y@g****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| d5356e07 |
Auto-detect CUTLASS for EvoformerAttention (#8000)
Co-authored-by: Masahiro Tanaka <m****a@a****m>, Masahiro Tanaka <8****a@u****m> Signed-off-by: Max Tretikov <m****x@t****m>, Masahiro Tanaka <m****a@a****m> |
Max Tretikov <6****v@u****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| b01a0915 |
fix(io): close aio_fd in FastFileWriter._fini to prevent fd leak (#8005)
Co-authored-by: Masahiro Tanaka <8****a@u****m> Signed-off-by: jg-heo <c****o@g****m> |
jg-heo <c****o@g****m>
Committed by: GitHub <n****y@g****m> |
3 months ago | |
| 66af8f03 |
zero3: SDMA allgather via mori (sdma_allgather) (#7999)
Co-authored-by: wuyl1 <y****u@a****m>, wuyl1 <y****u@a****m> Signed-off-by: wuyl1 <y****u@a****m>, inkcherry <m****u@a****m> |
inkcherry <m****u@a****m>
Committed by: GitHub <n****y@g****m> |
4 months ago | |
| 55491413 |
Optimize singleton MoE collectives (#7997)
Co-authored-by: Ma, Guokai <g****a@g****m> Signed-off-by: Tianyi Wang <n****n@g****m> |
Tianyi Wang <n****n@g****m>
Committed by: GitHub <n****y@g****m> |
4 months ago | |
| 5be4c70c |
Update topkgating probability-mask test expectation (#8007)
Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
4 months ago | |
| 883fefb2 |
Add office hours times/link on the README (#8004)
Signed-off-by: Logan Adams <l****s@m****m> |
Logan Adams <1****s@u****m>
Committed by: GitHub <n****y@g****m> |
4 months ago | |
| a7811998 | Update version post-v0.19.0 release (#7996) |
Logan Adams <1****s@u****m>
Committed by: GitHub <n****y@g****m> |
4 months ago | |
| 98a2fbdf | Update version.txt before 0.19.0 release (#7995) |
Logan Adams <1****s@u****m>
Committed by: GitHub <n****y@g****m> |
4 months ago | |
| 4e668fce |
[AutoSP] (Sequence Parallelism) support for Multimodal Models (ViT + LLM) (#7...
Co-authored-by: copilot-swe-agent[bot] <1****t@u****m>, nathon-lee <2****e@u****m>, Ma, Guokai <g****a@g****m> Signed-off-by: nathon-lee <l****o@g****m> |
nathon <l****o@g****m>
Committed by: GitHub <n****y@g****m> |
4 months ago | |
| 8a77f381 |
Add Gram Newton-Schulz orthogonalization for Muon optimizer (#7953)
Co-authored-by: Zhipeng Wang <z****e@g****m> Signed-off-by: Ma, Guokai <g****a@g****m> |
Ma, Guokai <g****a@g****m>
Committed by: GitHub <n****y@g****m> |
4 months ago | |
| 5b441027 |
Add DeepSpeed NVTX domain support (#7988)
Co-authored-by: Olatunji Ruwase <t****e@s****m>, Ma, Guokai <g****a@g****m> Signed-off-by: heurry <r****2@1****m> |
heurry <1****y@u****m>
Committed by: GitHub <n****y@g****m> |
4 months ago | |
| 853c9389 |
fix: topkgating major bug (#7986)
Signed-off-by: Daniel Shen <d****2@g****m> |
Daniel Shen <5****l@u****m>
Committed by: GitHub <n****y@g****m> |
4 months ago | |
| 5999fb06 |
Fix BF16_Optimizer last-microbatch grad leak under ZeRO-1 (#7985)
Co-authored-by: Ma, Guokai <g****a@g****m>, Masahiro Tanaka <m****a@a****m> Signed-off-by: Max Yu <1****5@u****m>, Masahiro Tanaka <m****a@a****m> |
Max Yu <1****5@u****m>
Committed by: GitHub <n****y@g****m> |
4 months ago | |
| 6e291bee | Fix modal ci timeout (#7989) |
Olatunji Ruwase <t****e@s****m>
Committed by: GitHub <n****y@g****m> |
4 months ago | |
| f6d7ae48 |
Dynamic offload compatible with static optimizer offload (#7979)
Co-authored-by: Masahiro Tanaka <8****a@u****m> |
Olatunji Ruwase <t****e@s****m>
Committed by: GitHub <n****y@g****m> |
4 months ago | |
| 44c51e34 |
Fix Adam subgroup inconsistency (#7982)
Signed-off-by: st_bang <s****g@d****r> |
st-bang97 <s****g@d****r>
Committed by: GitHub <n****y@g****m> |
4 months ago | |
| 077bff55 |
deepcompile: Fix backward graph recompilation due to unbalanced forward/backw...
Signed-off-by: Junjie Mao <j****o@l****m> |
Junjie Mao <j****o@l****m>
Committed by: GitHub <n****y@g****m> |
4 months ago | |
| aeb10bb1 |
Fix ZeRO-1/2 CPU-offloaded gradient loss with multiple backward() per step (#...
Signed-off-by: Sung Hyun Cho <h****7@g****m> |
Sung Hyun Cho <h****7@g****m>
Committed by: GitHub <n****y@g****m> |
4 months ago | |
| 83d43c35 |
Avoid CUDA reinit error in CI tests (#7977)
Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
4 months ago | |
| dc0fd295 |
Rename dequantization template parameters (#7976)
Signed-off-by: Alexander Grund <a****d@t****e> |
Alexander Grund <F****e@u****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 893c6d20 |
fix(op_builder): avoid duplicate/wrong -gencode flags (#7974)
Signed-off-by: Cursx <6****1@q****m> |
Cursx <3****x@u****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 3ba3dc97 |
fix(fp_quantizer): fix UB and negative shift warnings in fp_quantize_impl.cu ...
Signed-off-by: Cursx <6****1@q****m> |
Cursx <3****x@u****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 0ba23529 |
Enable PyTorch version selection for full test (#7968)
Co-authored-by: Ma, Guokai <g****a@g****m> Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 3fd762cf |
Fix WarmupCosineLR multi-group initialization (#7969)
Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| ecb26a51 |
Fix DeepCompile+Z3 on PyTorch v2.9/2.10 (#7951)
Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| dac1525b |
Fix overlap-comm buffer lifetimes (#7965)
Co-authored-by: Ma, Guokai <g****a@g****m> Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 9d632f10 |
Fix zero 1 and 2 CPU-offloaded gradient norm (#7967)
Signed-off-by: Alexis Limozin <a****s@l****t> |
Alexis Limozin <1****n@u****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 3a5df512 |
Fix FPQuantizer build (#7963)
Signed-off-by: Alexander Grund <a****d@t****e> |
Alexander Grund <F****e@u****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 37e232fa |
fix(zero): detach flat buffer to prevent autograd inplace error on CP… (#7948)
Signed-off-by: Guokai Ma <g****a@i****m>, Ma, Guokai <g****a@g****m> |
Ma, Guokai <g****a@g****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| bf0126b5 |
[SP] add SP deny list instead of allow (#7887)
Co-authored-by: Masahiro Tanaka <8****a@u****m>, Logan Adams <1****s@u****m>, Olatunji Ruwase <t****e@s****m>, Stas Bekman <s****0@u****m> Signed-off-by: Kashif Rasul <k****l@g****m> |
Kashif Rasul <k****l@g****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 5dce124e |
Zero3 defragment utility (#7940)
Co-authored-by: copilot-swe-agent[bot] <1****t@u****m>, nathon-lee <2****e@u****m> Signed-off-by: nathon-lee <l****o@g****m> |
nathon <l****o@g****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 2f0924a5 |
Fix process hang in process-group shutdown (#7941)
Signed-off-by: Alexander Grund <a****d@t****e> |
Alexander Grund <F****e@u****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 3bdebc03 |
Fix/fix autotp universal checkpoint ci (#7937)
Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 89bf0d2d |
Refactor consolidate transpose (#7934)
Co-authored-by: copilot-swe-agent[bot] <1****t@u****m>, nathon-lee <2****e@u****m>, Olatunji Ruwase <t****e@s****m> Signed-off-by: nathon-lee <l****o@g****m> |
nathon <l****o@g****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 607b55f1 |
Update version after latest release (v0.18.9) (#7936)
Signed-off-by: Logan Adams <l****s@m****m> |
Logan Adams <1****s@u****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 8c93851e | Update version.txt for latest incoming release 0.18.9 (#7935) |
Logan Adams <1****s@u****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 9486ab75 |
Remove Microsoft Corporation copyright from AGENTS.md and CLAUDE.md (#7932)
Co-authored-by: Zhipeng Wang <z****a@w****u>, Olatunji Ruwase <t****e@s****m> Signed-off-by: Zhipeng Wang <z****a@w****u> |
Zhipeng Wang <z****e@g****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 36f0b0c7 |
Add fallback to full test (#7933)
Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 5efb24ac |
Merging AutoSP into DeepSpeed (#7860)
Co-authored-by: Ahan Gupta <a****6@g****m>, Olatunji Ruwase <t****e@s****m>, Masahiro Tanaka <8****a@u****m>, Zhipeng Wang <z****e@g****m> Signed-off-by: Neel Dani <n****8@g****m>, Ahan Gupta <a****6@g****m>, Masahiro Tanaka <m****a@a****m> |
Neel Dani <n****8@g****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 2bae3606 | Update README with latest news from DeepSpeed (#7931) |
Zhipeng Wang <z****e@g****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| abb88ceb |
AGENTS.md: Add pre-commit command to existing CI requirements line (#7930)
Signed-off-by: Guokai Ma <g****a@i****m> |
Ma, Guokai <g****a@g****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 729df6ca |
fix(superoffload) preserve multi-group updates with shared cpu buffers (#7905...
Co-authored-by: Olatunji Ruwase <t****e@s****m> |
Xinyu Lian <l****7@i****u>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 62c3e6d8 | Add news item for ASPLOS 2026 Best Paper Award (#7923) |
Zhipeng Wang <z****e@g****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 956ec6fb |
Extending Muon Optimizer Support for ZeRO Stage 3 (#7919)
Co-authored-by: Claude Opus 4.6 (1M context) <n****y@a****m> Signed-off-by: PKUWZP <z****e@g****m> |
Zhipeng Wang <z****e@g****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 138f20d9 |
Fix backward compatibility of torch.amp.custom_fwd for PyTorch < 2.4 (#7920)
Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| f887b982 |
fix: handle non-existent path in is_nfs_path for Triton autotune cache (#7921)
Co-authored-by: Masahiro Tanaka <8****a@u****m> Signed-off-by: Krishna Chaitanya Balusu <k****5@g****m> |
Krishna Chaitanya <k****5@g****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| a240c4da |
Add HuggingFace tp_plan support for AutoTP (#7901)
Co-authored-by: Olatunji Ruwase <t****e@s****m>, Masahiro Tanaka <8****a@u****m> Signed-off-by: Guokai Ma <g****a@i****m>, Ma, Guokai <g****a@g****m> |
Ma, Guokai <g****a@g****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 26c954ff |
Don't detect local GPU if `$DS_IGNORE_CUDA_DETECTION` is set (#7896)
Co-authored-by: Olatunji Ruwase <t****e@s****m>, Masahiro Tanaka <8****a@u****m> |
Alexander Grund <F****e@u****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 5ce3abbf |
fix: remove unnecessary shell=True in ROCm GPU architecture detection (#7915)
Co-authored-by: Masahiro Tanaka <8****a@u****m> |
instantraaamen <1****n@u****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| f2bb1ec6 |
Add Feature Universal Checkpoint for AutoTP (#7908)
Co-authored-by: copilot-swe-agent[bot] <1****t@u****m>, nathon-lee <2****e@u****m>, Ma, Guokai <g****a@g****m> Signed-off-by: nathon-lee <l****o@g****m> |
nathon <l****o@g****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 38bd11a7 |
Respect `$TRITON_HOME` (#7907)
Signed-off-by: Alexander Grund <a****d@t****e> |
Alexander Grund <F****e@u****m>
Committed by: GitHub <n****y@g****m> |
5 months ago | |
| 5f7b6870 | Update version (#7903) |
Logan Adams <1****s@u****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| be60451f |
fix(zero3): use current_stream() instead of default_stream() for grad… (#7898)
Co-authored-by: Olatunji Ruwase <t****e@s****m> Signed-off-by: Ubuntu <u****u@i****l> |
Michael Royzen <4****n@u****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| 49a82a0d |
Add AGENTS.md and CLAUDE.md with project rules for AI coding agents (#7902)
Signed-off-by: Ma, Guokai <g****a@g****m> |
Ma, Guokai <g****a@g****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| 63eeb114 |
fix: Validate fp16.loss_scale is finite and non-negative (#7889)
Co-authored-by: Masahiro Tanaka <8****a@u****m> Signed-off-by: nathon-lee <l****o@g****m> |
nathon <l****o@g****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| 784cc26e |
Fix Evoformer's multi-arch dispatch root cause (#7881)
Co-authored-by: Olatunji Ruwase <t****e@s****m> Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| f88d0f8d |
Fix async_io ops building error on Huawei Ascend NPU (#7894)
Co-authored-by: Olatunji Ruwase <t****e@s****m> Signed-off-by: Huang Yifan <y****0@f****m> |
Huang Yifan <y****0@f****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| b6346bf0 | double reduction user-friendly error (#7895) |
Stas Bekman <s****0@u****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| d9a4aadb |
[Bloom] Fix hangs of bloom test (#7890)
Co-authored-by: Olatunji Ruwase <t****e@s****m> Signed-off-by: Artem Kuzmitckii <a****i@a****m> |
Artem Kuzmitckii <a****i@a****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| 285cae30 |
Suppress see_memory_usage logs (#7891)
Signed-off-by: Olatunji Ruwase <t****e@s****m> |
Olatunji Ruwase <t****e@s****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| 6c59d544 |
Fix hook count performance regression from v0.18.5 (#7886)
Co-authored-by: Ramya Ramineni <r****n@u****m> Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| 4dba1e20 |
Add document section explaining autocast nesting (#7883)
Co-authored-by: Stas Bekman <s****0@u****m> Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| a41a96b1 |
Remove amp() from abstract accelerator (#7879)
Co-authored-by: Olatunji Ruwase <t****e@g****m> Signed-off-by: Ma, Guokai <g****a@i****m> |
Ma, Guokai <g****a@g****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| d8e15da4 |
XPU use stock pytorch instead of Intel Extension for PyTorch (#7877)
Co-authored-by: Olatunji Ruwase <t****e@g****m>, Olatunji Ruwase <t****e@s****m> Signed-off-by: Ma, Guokai <g****a@i****m> |
Ma, Guokai <g****a@g****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| 04d69cc6 |
Fix import deepspeed crash on PyTorch v2.3 + Python 3.12 (#7875)
Co-authored-by: Olatunji Ruwase <t****e@g****m> Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| bffaf457 |
Fix no-grad grad-fn lookup in ZeRO hook counting on PyTorch 2.3 (#7830) (#7841)
Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| 116dbe28 |
Fix: only add parameter with grads to parameter group (#7869)
Signed-off-by: Ma, Guokai <g****a@i****m> |
Ma, Guokai <g****a@g****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| a15e557d |
perf: use deque for FIFO queues in sequence parallel, superoffload, and compi...
Signed-off-by: g97iulio1609 <g****e@g****m> |
Giulio Leone <g****e@g****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| ae21699a |
fix: replace 35 bare except clauses with except Exception (#7873)
Co-authored-by: haosenwang1018 <h****8@u****m> |
Sense_wang <1****8@u****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| efc0b49a |
Fix broken links and add AutoTP Training tutorial to sidebar nav (#7874)
Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| 0416cf68 |
Schedule nightly full test (#7870)
Co-authored-by: Logan Adams <1****s@u****m> Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| 93524c89 |
Fix Zero-3 static scale assertion in fp16 test (#7866)
Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| 57b10d54 |
Throw error when parameter is modified in GatheredParameters (#7832)
Co-authored-by: Olatunji Ruwase <t****e@s****m> Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| c9e652cf |
Fix compilation of Evoformer (#7862)
Signed-off-by: Alexander Grund <a****d@t****e> |
Alexander Grund <F****e@u****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| dbc1b075 |
Fix ROCm BF16 conversion intrinsics in inference v2 (#7843) (#7846)
Signed-off-by: Masahiro Tanaka <m****a@a****m> |
Masahiro Tanaka <8****a@u****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| f3a9819c |
Add EXAONE 4.0 model support for Inference V2 (#7853)
Signed-off-by: Bias92 <p****5@g****m> |
Bias29 <8****2@u****m>
Committed by: GitHub <n****y@g****m> |
6 months ago | |
| 8a63a2a9 | Add news entry for DeepSpeed updates (#7854) |
Zhipeng Wang <z****e@g****m>
Committed by: GitHub <n****y@g****m> |
6 months ago |