MolmoAct 2
Allen Institute for AIthe loss is computed only on the action tokens
implementation detail · filed under training objective
Computes cross-entropy supervision only on predicted action tokens, rather than on all sequence tokens.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See language modelling objective for the methods filed beside it.
Also called Action-token-only next-token training loss, loss is computed only on the action tokens, standard next-token prediction objective.
3 quotes from 2 sources, one card per model, each opening with its strongest.
the loss is computed only on the action tokens
OpenVLA is trained with a standard next-token prediction objective, evaluating the cross-entropy loss on the predicted action tokens only
evaluating the cross-entropy loss on the predicted action tokens only