MolmoAct 2
Allen Institute for AIOverall, full-model adaptation with discrete and continuous action co-training provides the best average outcome.
specific method · filed under post-training
Updates all model weights during adaptation rather than freezing selected components.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See supervised fine-tuning for the methods filed beside it.
Also called full-model adaptation.
3 quotes from 3 sources, one card per model, each opening with its strongest.
Overall, full-model adaptation with discrete and continuous action co-training provides the best average outcome.
We fine-tune the VLA models for robot action prediction, without weight freezing.
full fine-tuning of all model parameters, using small datasets with 10–150 demonstrations of a target task