LAW
Collection
LAW reproduction resources: four final models, three training initializations, and two complete training datasets. • 9 items • Updated
Complete Q1/Q2/Q3 VLM for LAW-OpenWAM: task planning, completion/failure assessment and action-value estimation. Includes the backbone, adapters and tokenizer; Q3 is under q3/.
Follow the LAW README for setup, training and policy inference. Download from the LAW repository:
"$PYTHON_BIN" scripts/download.py models --profile 20d
LAW-trained weights: Apache-2.0. Upstream components retain their terms; see license scope.