5 Pull requests merged by 5 people

• hf tp+zero training doc.

  #7151 merged Mar 21, 2025

• Update container version that runs on A6000 tests.

  #7153 merged Mar 20, 2025

• Enhance Gaudi2 CI/Nightly Coverage with Model Parallelism and Linear Tests

  #7146 merged Mar 19, 2025

• Correct the BACKWARD_PREFETCH_SUBMIT mismatch

  #7120 merged Mar 17, 2025

• Conditionally quote env vars

  #7071 merged Mar 17, 2025

8 Pull requests opened by 7 people

• [NFC] Typo fix in SP layer.

  #7152 opened Mar 19, 2025

• DeepCompile for enhanced compiler integration

  #7154 opened Mar 19, 2025

• Avoid graph break by removing redundant requires_grad attr change

  #7158 opened Mar 20, 2025

• Add destroy to tests to free memory

  #7160 opened Mar 20, 2025

• fixed: Modified the topkgating function and modified the test_moe file for testing

  #7163 opened Mar 21, 2025

• Add DataStates-LLM: Asynchronous Checkpointing Engine Support

  #7166 opened Mar 21, 2025

• Link AutoTP blog in the front page

  #7167 opened Mar 21, 2025

• Fix pre-compile on cpu-only machines

  #7168 opened Mar 22, 2025

3 Issues closed by 3 people

• [BUG] `PydanticDeprecatedSince20` coming from `DeepSpeedZeroConfig`

  #7149 closed Mar 21, 2025

• Does DeeepSpeed support autotp and pp in llm training?

  #7147 closed Mar 19, 2025

• [BUG]error: #error "You're trying to build PyTorch with a too old version of GCC. We need GCC 9 or later."

  #7143 closed Mar 18, 2025

12 Issues opened by 12 people

• nv-ds-chat CI test failure

  #7169 opened Mar 23, 2025

• Install DeepSpeed fail with setuptools-77.0.3

  #7165 opened Mar 21, 2025

• [BUG] Enabling hpZ causes an abnormally large loss.

  #7164 opened Mar 21, 2025

• Add DCO check guidance.

  #7162 opened Mar 21, 2025

• [BUG] circular import on `DeepSpeedTransformerInference`

  #7159 opened Mar 20, 2025

• [BUG] AttributeError: module 'deepspeed' has no attribute 'init_inference'

  #7157 opened Mar 20, 2025

• [REQUEST] Support for Expert Optimizer State Partitioning with ZeRO Optimization in DeepSpeed MoE

  #7156 opened Mar 20, 2025

• no slot '4' specified on local host - trying to use 4 gpus on a node with 8 gpus while another user is using the other 4 gpus

  #7155 opened Mar 19, 2025

• [BUG] Receiving CUDA error: invalid argument using pytorch 2.7 with deepspeed 0.16.4 with Cuda 12.8

  #7150 opened Mar 19, 2025

• [REQUEST]Does the current version support distributed fine-tuning on mac devices (M2-M4)?

  #7148 opened Mar 18, 2025

• static loss scale with stage 0 occured error

  #7145 opened Mar 17, 2025

• [REQUEST] Support for Nvidia 50 Series GPUs: Pytorch >=2.6 and CUDA 12.8 required

  #7144 opened Mar 17, 2025

17 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Enable torch.autocast with ZeRO

  #6993 commented on Mar 21, 2025 • 2 new comments

• async tp allreduce

  #7115 commented on Mar 21, 2025 • 1 new comment

• [XPU] Support XCCL on deepspeed side

  #7113 commented on Mar 17, 2025 • 0 new comments

• Variable batch size and LR scheduler

  #7104 commented on Mar 21, 2025 • 0 new comments

• Unpin once transformers latest is fixed

  #7088 commented on Mar 20, 2025 • 0 new comments

• Enable ZeRO set/get APIs for NVMe offload

  #7046 commented on Mar 21, 2025 • 0 new comments

• Add `pyproject.toml` with legacy build backend to keep most logic in `setup.py`

  #7033 commented on Mar 20, 2025 • 0 new comments

• Improve overflow handling in ZeRO

  #6976 commented on Mar 21, 2025 • 0 new comments

• Fix: forbid repeated deepspeed.initialize on training objects

  #6874 commented on Mar 21, 2025 • 0 new comments

• [BUG] Zero2 offload overflow

  #5241 commented on Mar 23, 2025 • 0 new comments

• nv-nightly CI test failure

  #7140 commented on Mar 23, 2025 • 0 new comments

• safe_get_full_grad & safe_set_full_grad

  #7117 commented on Mar 21, 2025 • 0 new comments

• [BUG] DeepSpeed accuracy issue for torch.compile if activation checkpoint function not compiler disabled

  #6718 commented on Mar 21, 2025 • 0 new comments

• [BUG] Batch inference DDP + zero stage 3 = inference code hangs

  #7128 commented on Mar 20, 2025 • 0 new comments

• [BUG] OOM when train 70B models using deepspeed 0.16.4

  #7116 commented on Mar 19, 2025 • 0 new comments

• [REQUEST] torch.compile + DeepSpeed

  #4677 commented on Mar 19, 2025 • 0 new comments

• [BUG?] zero++ init problem

  #7066 commented on Mar 18, 2025 • 0 new comments