single mention · not yet filed
Per-layer key-value conditioning of a flow-matching action expert
Per-layer key-value conditioning lets a flow-matching action expert receive VLM context at every layer rather than from a single final hidden state.1 Each action-expert block cross-attends to the corresponding VLM layer's keys and values, mapped by learned projections into the expert's cross-attention width.2
Trade-off It outperforms hidden-state conditioning on average, though the passages report no cost.3
How this definition was written
Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.
MolmoAct2 conditions the expert on the VLM at every layer
— MolmoAct2 Action Reasoning Models for Real-World DeploymentEach action-expert block cross-attends to the corresponding VLM layer’s keys and values, after lightweight learned projections map the VLM attention state into the expert’s cross-attention width
— MolmoAct2 Action Reasoning Models for Real-World DeploymentThe standard per-layer KV conditioning is strongest on average, reaching 95.9
— MolmoAct2 Action Reasoning Models for Real-World Deployment
Also called per-layer KV conditioning.