Skip to content

feat: algorithm abstraction — named algorithm classes + inline frozen-model references (grpo, opd, sft_distill, self_distill, echo)#2746

Open
hallerite wants to merge 47 commits into
mainfrom
feat/algorithm-abstraction
Open

feat: algorithm abstraction — named algorithm classes + inline frozen-model references (grpo, opd, sft_distill, self_distill, echo)#2746
hallerite wants to merge 47 commits into
mainfrom
feat/algorithm-abstraction

fix(trainer): namespace ref_kl loss metrics; harden batch preparation

0606f12
Select commit
Loading
Failed to load commit list.