Search results

[BUG] IndexError in BF16_Optimizer.destroy() when optimizer=None (DummyOptim)

bugSomething isn't workingSomething isn't working

inference

Status: Open.

#7752 In deepspeedai/DeepSpeed;

· ptx9363 opened on Dec 30, 2025

[BUG] Grad norm value differs significantly across DeepSpeed version

bugSomething isn't workingSomething isn't working

training

Status: Open.

#7749 In deepspeedai/DeepSpeed;

· npurson opened on Dec 29, 2025

[BUG]MoE router parameters are forced to bf16 under DeepSpeed bf16, causing dtype mismatch in fp32 routing logic

bugSomething isn't workingSomething isn't working

training

Status: Open.

#7747 In deepspeedai/DeepSpeed;

· tangefly opened on Dec 27, 2025

[BUG] Dtype mismatch (bf16 vs fp32) when resuming Muon optimizer from checkpoint

bugSomething isn't workingSomething isn't working

training

Status: Open.

#7746 In deepspeedai/DeepSpeed;

· literid opened on Dec 25, 2025

Please publish DeeepSpeed Python 3.13 Windows wheels already

ci-failure

Status: Open.

#7744 In deepspeedai/DeepSpeed;

· FurkanGozukara opened on Dec 23, 2025

Deepspeed dual-machine single-card streaming inference has an error

Status: Open.

#7734 In deepspeedai/DeepSpeed;

· Haneeeeeeee opened on Dec 18, 2025

[BUG] The different lr for Muon doesn't work?

bugSomething isn't workingSomething isn't working

training

Status: Open.

#7713 In deepspeedai/DeepSpeed;

· Zazexy opened on Dec 4, 2025

[BUG]

bugSomething isn't workingSomething isn't working

training

Status: Open.

#7709 In deepspeedai/DeepSpeed;

· Aurora2025-tech opened on Nov 28, 2025

[Roadmap] DeepSpeed Roadmap Q1 2026

Status: Open.

#7705 In deepspeedai/DeepSpeed;

· PKUWZP opened on Nov 26, 2025

How to use tp or tp when I traning models from transformers?

Status: Open.

#7702 In deepspeedai/DeepSpeed;

· Mulbetty opened on Nov 24, 2025

ignoring *.cuh prevents multi_tensor_apply.cuh from being pushed

Status: Open.

#7701 In deepspeedai/DeepSpeed;

· LckyLke opened on Nov 23, 2025

[BUG] Significant Performance Difference between DeepSpeed's zero_stage=1 and zero_stage=2

bugSomething isn't workingSomething isn't working

training

Status: Open.

#7697 In deepspeedai/DeepSpeed;

· liyc-ai opened on Nov 18, 2025