deepspeedai/DeepSpeed

Pull Requests Commits

Add Triton grouped-GEMM for MoE experts on Ampere/Ada (#8180)

hwchen2017 committed 13 hours ago

Verified df84f6d8

Preserve tensor learning rates in OneCycle (#8205)

tohtana committed 17 hours ago

Verified f6a386a3

Remove redundant DCO workflow (#8188)

tohtana committed 20 hours ago

Verified d2bd087c

Add keyword argument support to activation checkpointing (#8182)

zupengwang committed 21 hours ago

Verified ebf1531e

MoE dispatch: fix silent Tutel + TP corruption, speed up native and Tutel paths (#8195)

jinyouzhi committed 1 day ago

Verified 619e3199

Add managed_gradient_accumulation for ZeRO stage 0/1 (#8184)

sfc-gh-truwase committed 1 day ago

Verified 7a31fe66

[AutoTP] Fix ZeRO-3 checkpoint consolidation to gather across TP and DP (#8168)

delock committed 1 day ago

Verified eec237ee

Preserve tensor learning rates across scheduler updates (#8202)

n33levo committed 1 day ago

Verified 3b641487

Clamp capacity to num_tokens in MoE gating drop branches (complete #5353) (#8155)

ebarkhordar committed 1 day ago

Verified c43cf21c

Skip CUDA transformer forward tests when TransformerBuilder is incompatible (#8200)

sfc-gh-truwase committed 2 days ago

Verified 7e07be2a

Enable ZeRO-3 linear wrapper for existing models (#8189)

tohtana committed 2 days ago

Verified 617061d6

Release GIL in blocking AIO bindings (#8196)

tohtana committed 3 days ago

Verified e2aae1b0

Warn when zero.Init silently falls back to a single rank (#8084) (#8089)

akshansh47 committed 3 days ago

Verified a375b7f1

Derive AutoEP rank splits from the per-expert count exchange (#8190)

yh0903 committed 3 days ago

Verified 35d1c7d3

[AutoTP] Preserve universal checkpoint metadata for HuggingFace tp_plan (#8187)

jinyouzhi committed 4 days ago

Verified d6fbd476

Clarify merge commit exemption for signoff check in AGENTS.md and CLAUDE.md (#8194)

jinyouzhi committed 4 days ago

Verified 5cc06170

Scope DeepCompile compiler state to graph and engine lifecycles (#8159)

tohtana committed 4 days ago

Verified 615e6d5e

[AutoTP] Enable HF `colwise_gather_output` to support `lm_head` replace (#8146)

jinyouzhi committed 5 days ago

Verified 48d54e05

Make the OneCycle stair counts actually change the schedule (#8179)

vineethsaivs committed 5 days ago

Verified 3169c2dc

Fix WarmupLR collapsing multi-group base LRs to group 0's (#8171)

ebarkhordar committed 5 days ago

Verified 6c58fb75

Guard LRRangeTest and OneCycle schedulers against zero step sizes (#8166)

ebarkhordar committed 5 days ago

Verified 90e30f4d

Stabilize fork-sensitive nightly tests and AutoSP coverage (#8172)

tohtana committed 5 days ago

Verified 86fd223f

NVMe write warning (#8164)

sfc-gh-truwase committed 7 days ago

Verified 010c6e91

Enable support for Tutel when k != 1 for shared moe (#8174)

jinyouzhi committed 7 days ago

Verified dc81c9ee

Fix autotuning get_val_by_key to search all nested subdicts (#8177)

ebarkhordar committed 7 days ago

Verified 4b693658

[AutoTP] Allow ZeRO stage 3 inference with tensor parallelism (#8167)

delock committed 9 days ago

Verified d3265209

Run pull request code only inside a Modal Sandbox (#8170)

tohtana committed 10 days ago

Verified 39adc2b1

Validate warmup_type in WarmupCosineLR like WarmupLR (#8151)

sohumt123 committed 11 days ago

Verified 886790b5

Adopt PTF Code of Conduct (#8165)

sfc-gh-truwase committed 11 days ago

Verified 0c36f6d3

Z3: Support for activation ckpt with frozen params (#8148)

sfc-gh-truwase committed 12 days ago

Verified a31eb316

Older