MolmoAct2-SO101-Mdrift_vtptoponly60-D6231-step51000

A user-published fine-tuned checkpoint derived from Ai2's MolmoAct2 vision-language-action model family, adapted for SO-101-style robot manipulation tasks. It extends the base MolmoAct2 architecture — which combines image/video perception with continuous action prediction — via task-specific fine-tuning at training step 51,000. Source: supabase.