分支列表

默认分支

71d316d608 · Deprecate unused DeepSpeed features (#8490) · 最后更新于 2026-09-11 22:50:44 +00:00

分支列表

fb731ff9eb · Initial plan · 最后更新于 2026-07-01 18:50:17 +00:00    frostbyte_neo

198
1

6f4a7afb6f · Fix eigenvalue monitoring: list() wrapper and correct variable name · 最后更新于 2026-07-01 17:34:55 +00:00    frostbyte_neo

198
2

918b69043b · fix(rollout): fix CUDA graph capture producing incorrect logits · 最后更新于 2026-06-08 15:08:58 +00:00    frostbyte_neo

313
32

aa01593168 · move trainer to deepspeed/runtime/rlhf/trainer · 最后更新于 2026-06-03 09:28:34 +00:00    frostbyte_neo

313
15

ab9e956c2d · Preserve selected Transformers in torch latest CI · 最后更新于 2026-05-20 04:51:21 +00:00    frostbyte_neo

260
11

f060338fe8 · test: mark FastFileWriter fd tests sequential · 最后更新于 2026-05-20 03:52:11 +00:00    frostbyte_neo

262
3

00edd290f6 · Fix AutoEP + Muon compatibility for batched expert tensors · 最后更新于 2026-05-19 06:52:37 +00:00    frostbyte_neo

266
82

e143244dba · [Ulysses SP] Add SP-aware Qwen3.5 GatedDeltaNet forward · 最后更新于 2026-05-11 04:23:48 +00:00    frostbyte_neo

269
1

809c56c98c · docs: add test directory convention to AGENTS.md · 最后更新于 2026-05-06 07:00:02 +00:00    frostbyte_neo

271
1

088fd97c43 · autotp: fix qwen3.5 linear attention tp · 最后更新于 2026-04-16 12:23:17 +00:00    frostbyte_neo

282
3

7f8aff5120 · test(zero): add training step coverage for accelerator flatten path · 最后更新于 2026-04-03 00:14:25 +00:00    frostbyte_neo

303
3

b14e5b4616 · Address tohtana's review feedback for XLA/TPU support · 最后更新于 2026-03-26 07:38:55 +00:00    frostbyte_neo

314
9

da9b77384a · Address review feedback for Muon ZeRO Stage 3 support · 最后更新于 2026-03-26 00:08:43 +00:00    frostbyte_neo

314
2

169c828af6 · Add Feature Universal Checkpoint for AutoTP (#7908) · 最后更新于 2026-03-23 10:07:19 +00:00    frostbyte_neo

314
1

15aa9fb0bb · Merge 1ea484bc4a into a44fb58f13 · 最后更新于 2026-02-09 14:57:26 +00:00    frostbyte_neo

347
7

040316e973 · show warning when multiple matches found · 最后更新于 2026-01-30 22:48:17 +00:00    frostbyte_neo

362
5

f8998a86bd · Add Muon Optimizer Support to ZeRO3 · 最后更新于 2026-01-20 01:18:51 +00:00    frostbyte_neo

368
1

dca165bf5e · Add CUTLASS support and mark Evoformer test as sequential · 最后更新于 2026-01-18 22:52:39 +00:00    frostbyte_neo

375
47

97e7430161 · fix: skip compressed allreduce for empty tensors (#7769) · 最后更新于 2026-01-15 22:18:10 +00:00    frostbyte_neo

375
0
已包含

87f1979d9f · feat: Add Triton-optimized Muon optimizer implementation · 最后更新于 2025-11-14 00:47:52 +00:00    frostbyte_neo

448
4