What's Changed

• Update version.txt after release by @loadams in https://github.com/deepspeedai/DeepSpeed/pull/7675
• [modal ci] fixes by @stas00 in https://github.com/deepspeedai/DeepSpeed/pull/7676
• leaf modules: explain better by @stas00 in https://github.com/deepspeedai/DeepSpeed/pull/7674
• disable nv-lightning-v100.yml cI by @stas00 in https://github.com/deepspeedai/DeepSpeed/pull/7681
• allow seperate learning rate "muon_lr" and "adam_lr" for muon optimizer by @delock in https://github.com/deepspeedai/DeepSpeed/pull/7658
• see_mem_usage: make always work by @stas00 in https://github.com/deepspeedai/DeepSpeed/pull/7688
• make debug utils more resilient by @stas00 in https://github.com/deepspeedai/DeepSpeed/pull/7690
• zero stage 1-2: don't pin memory if not configured by @stas00 in https://github.com/deepspeedai/DeepSpeed/pull/7689
• modal ci: fix group concurrency by @stas00 in https://github.com/deepspeedai/DeepSpeed/pull/7691
• Use pytorch utils to detect ninja by @Emrys-Merlin in https://github.com/deepspeedai/DeepSpeed/pull/7687
• Update SECURITY.md to point to GitHub reporting rather than Microsoft by @loadams in https://github.com/deepspeedai/DeepSpeed/pull/7692
• Add Qwen2.5 to AutoTP model list by @delock in https://github.com/deepspeedai/DeepSpeed/pull/7696
• Trust intel server for XPU tests by @tohtana in https://github.com/deepspeedai/DeepSpeed/pull/7698
• PyTorch-compatible backward API by @tohtana in https://github.com/deepspeedai/DeepSpeed/pull/7665
• Add news about Ray x DeepSpeed Meetup by @PKUWZP in https://github.com/deepspeedai/DeepSpeed/pull/7704
• Put Muon optimizer momentum buffer on GPU by @delock in https://github.com/deepspeedai/DeepSpeed/pull/7648
• [ROCm] Relax tolerances for FP8 unit test for fp16 and bf16 cases by @rraminen in https://github.com/deepspeedai/DeepSpeed/pull/7655
• Fix that ds_secondary_tensor may be dirty when loading the model or zero checkpoint for zero++. by @zhengchenyu in https://github.com/deepspeedai/DeepSpeed/pull/7707
• fix: skip aio wait when swap tensors is empty by @xylian86 in https://github.com/deepspeedai/DeepSpeed/pull/7712
• Low-precision master params/grads/optimizer states by @tohtana in https://github.com/deepspeedai/DeepSpeed/pull/7700
• Enabled compiled autograd for backward pass by @deepcharm in https://github.com/deepspeedai/DeepSpeed/pull/7667
• Wall clock timers API by @sfc-gh-truwase in https://github.com/deepspeedai/DeepSpeed/pull/7714

New Contributors

• @Emrys-Merlin made their first contribution in https://github.com/deepspeedai/DeepSpeed/pull/7687

Full Changelog: https://github.com/deepspeedai/DeepSpeed/compare/v0.18.2...v0.18.3