All issues

Search results

[REQUEST] Support FusedLion / FusedLamb and Zero++ quantized communication (hpZ / qgZ) on NPU

enhancementNew feature or requestNew feature or request

Status: Open.

#8626 In deepspeedai/DeepSpeed;

· VenusTZZ opened on Sep 22, 2026

Universal checkpoint conversion only knows Adam's state keys, so a Muon checkpoint cannot be converted

Status: Open.

#8608 In deepspeedai/DeepSpeed;

· alanhuangyoo opened on Sep 20, 2026

[BUG]_prepare_fp32_grad_for_sub_group clears every parameter's gradient once per sub-group

bugSomething isn't workingSomething isn't working

training

Status: Open.

#8586 In deepspeedai/DeepSpeed;

· pengdurice opened on Sep 18, 2026

init_inference rejects dtype=torch.int8 on every accelerator

Status: Open.

#8577 In deepspeedai/DeepSpeed;

· udsy19 opened on Sep 18, 2026

[BUG] Runtime configuration validation is bypassed under python -O

Status: Open.

#8572 In deepspeedai/DeepSpeed;

· Saisharathchandranandnetha opened on Sep 17, 2026

2 tests fail: assert isinstance(model.llm[0], DistributedAttention)

Status: Open.

#8554 In deepspeedai/DeepSpeed;

· yurivict opened on Sep 16, 2026

[BUG] AutoTP checkpoint load raises AttributeError on a model containing nn.InstanceNorm1d

Status: Open.

#8514 In deepspeedai/DeepSpeed;

· ebarkhordar opened on Sep 14, 2026

[BUG] Python checkpoint writers fail to reload full ZeRO-3 state with default torch.load

Status: Open.

#8500 In deepspeedai/DeepSpeed;

· ai-yang opened on Sep 13, 2026

[BUG] ZeRO-3 ignores checkpoint tag when load_optimizer_states=False

Status: Open.

#8499 In deepspeedai/DeepSpeed;

· ai-yang opened on Sep 13, 2026

Deprecate unused DeepSpeed features

Status: Open.

#8489 In deepspeedai/DeepSpeed;

· sfc-gh-truwase opened on Sep 11, 2026

[BUG] Memory leak in loss backward when enabled torch activation checkpoint & zero3 & single-model-multi-branch network

bugSomething isn't workingSomething isn't working

training

Status: Open.

#8481 In deepspeedai/DeepSpeed;

· supermeng opened on Sep 11, 2026

Can't install in Python 3.14.6 with GCC

Status: Open.

#8480 In deepspeedai/DeepSpeed;

· Clayblockunova opened on Sep 11, 2026