LeWAM-Simulation-Checkpoint
Collection
LeWAM v1 checkpoints for simulation tasks: OGBench scene, puzzle (3x3), cube, DexMimicGen drawer, and PushT. • 13 items • Updated • 1
LeWAM Tool Hang (task completion)
Task-completion checkpoint in the LeWAM v1 format. Files are stored at the repository root:
lewam_best.pt
lewam_config.json
Download both files into $STABLEWM_HOME/checkpoints/<run_name>/, then evaluate with the v1 code:
python scripts/eval_lewam.py --config-name toolhang policy=<run_name> eval.full_traj=true eval.task_only=true
Trained on the scene camera (pixels) of LeWAM/lewam-toolhang, which is also the eval dataset.
The trained weights are preserved exactly. Strict model loading, attention masks, encoder output, action prediction, and dynamics prediction were checked against the source checkpoint. No optimizer state is included.
SHA256 of lewam_best.pt: 0252daddd9aa19e858c20f64855de52a27e3ba52967223eacb9098c5b2835761.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support