main
Clean up
[test] Fix/fix testcase (#5770)
[misc] fit torch api upgradation and remove legecy import (#6093)
[Zerobubble] merge main. (#6142)
[pre-commit.ci] pre-commit autoupdate (#5572)
[hotfix] Fix examples no pad token & auto parallel codegen bug; (#5606)
[Distributed RLHF] Integration of PP (#6257)
[hotfix] fix lora load (#6231)
[cli] support run as module option (#6135)
[shardformer] support ep for deepseek v3 (#6185)
[Fix]: implement thread-safety singleton to avoid deadlock for very large-scale training scenarios (#5625)
[Feature] Distributed optimizers: Lamb, Galore, CAME and Adafactor (#5694)
[upgrade]Upgrade transformers (#6320)
[Device]Support npu (#6159)
[fp8] Merge feature/fp8_comm to main branch of Colossalai (#6016)
[hotfix] moe hybrid parallelism benchmark & follow-up fix (#6048)
[optim] hotfix adam load (#6146)
[shardformer] support pipeline for deepseek v3 and optimize lora save (#6188)
[fp8] add fallback and make compile option configurable (#6092)
[Ring Attention] Add more detailed references (#6294)
[fp8] support fp8 amp for hybrid parallel plugin (#5975)
[hotfix] improve compatibility (#6165)
[hotfix] fix zero optim save (#6191)
[devops] remove post commit ci (#5566)
[fp8] hotfix backward hook (#6053)