StarVLA Dual Fold Blanket V4 ARX30 60k Checkpoints

This repository contains two 30-action-chunk ProgressVLA checkpoints trained on dual_fold_blanket_v4 with the ARX absolute joint-position contract.

Files:

dinoquery_arx30_60k/steps_60000_pytorch_model.pt
tripreadout_arx30_60k/steps_60000_pytorch_model.pt

Each variant also includes config.yaml, config.json, dataset_statistics.json, and summary.jsonl for reproducibility.

Contract:

action_type: abs_qpos
action_mode: abs
action_dim: 14
state_dim: 14
action_horizon: 30
future_action_window_size: 29
data_mix: dual_fold_blanket_v4_30chunk

Note: the original Slurm jobs reached 60000 steps and saved final checkpoints, but Slurm marked them failed due to a post-training shell EOF issue in an older launcher. The checkpoint artifacts listed here are complete 60k saves.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support