rlinf_libero_vla

RLinf LIBERO VLA Archive

This archive stores reproducible RLinf/OpenVLA-OFT LIBERO training recipes, model artifacts, checkpoints, logs, and evaluation summaries.

Layout

This model archive is intentionally separate from the independent /media/david/HDD/training_recipe/ repository:

• models/: base VLA model artifacts.
• checkpoints/: distributed PPO checkpoints by training run and global step.
• results/: metrics, logs, and TensorBoard outputs by training run.
• runs/: raw logs and TensorBoard snapshots.
• /media/david/HDD/training_recipe/: one self-contained recipe directory per training run, containing only YAML, source revision, hyperparameters, and README.

The first archived run is the 4-GPU H20 task-3 PPO experiment completed on 2026-09-18. Its final checkpoint is under checkpoints/2026-09-17_task3_ppo_4gpu_h20_gpu0126/global_step_100/; its recipe is under the independent training_recipe directory.

Storage and GitHub guidance

The archive is stored on centre at /media/david/HDD/rlinf_libero_vla (the mounted filesystem is uppercase HDD; /media/david/hdd is not available). The recipe YAMLs and READMEs are suitable for a GitHub repository. The model and distributed checkpoints are large binary artifacts (about 29 GB and 59 GB); keep them in Git LFS or an artifact store rather than regular Git history. The checkpoint byte totals are recorded beside the checkpoints.