SAVRN
Search Contact SAVRN

Independent publisher

Zihan Zhao

UltraMar1ne

Models in Library1
Datasets in Library0
Models on Hugging Face3
Followers

Models

Anonymized checkpoint ladders for a study of safety behavior under RLVR self-play and SFT. Folders are codenames; the codename->setting mapping is withheld. (actor/modelworldsize.pt + config); use veRL modelmerger/convert2hf to load. step180 is HF safetensors.

Open weights