--- language: - en license: mit tags: - SmolVLA - LeRobot - robotics - imitation-learning - behavior-cloning - so101 pipeline_tag: reinforcement-learning library_name: lerobot base_model: - lerobot/smolvla_base --- # LeRobot SO101 SmolVLA task1-unknown_bs64_s20000 ## Summary This repository contains the final checkpoint for a SmolVLA policy fine-tune trained on `aswinkumar99/task1-unknown` for SO101 sponge pick-and-place experiments. Dataset meaning: Task 1: Single Sponge - No Distractors (random-locations / unknown). This SmolVLA policy is a fine-tune of `lerobot/smolvla_base`, as recorded by both the launch command (`--policy.path=lerobot/smolvla_base`) and the saved training config (`pretrained_path: lerobot/smolvla_base`). ## Training Setup - Dataset repo: `aswinkumar99/task1-unknown` - Local dataset root during training: `/home/riftuser/datasets_combined/aswinkumar99/task1-unknown` - Output directory during training: `/home/riftuser/outputs_matrix/smolvla/task1-unknown_bs64_s20000` - Batch size: `64` - Training steps: `20000` - Checkpoint save frequency: `5000` - Data loader workers: `8` - WandB project: `so101-layout-generalization` - GPU: `NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition` - Python: `CPython 3.12.13` - CUDA: `12.9` - Training start: `2026-04-24T00:24:53.023942+00:00` - Training end: `2026-04-24T02:49:03` - Approximate training duration: `2h 24m 9s` - Base model: `lerobot/smolvla_base` - Observation camera rename map: `{"observation.images.overhead": "observation.images.camera1", "observation.images.wrist": "observation.images.camera2"}` - Action chunk size: `50` - Action steps predicted: `50` ## Exact Training Command ```bash lerobot-train \ --dataset.repo_id=aswinkumar99/task1-unknown \ --dataset.root=/home/riftuser/datasets_combined/aswinkumar99/task1-unknown \ --dataset.video_backend=torchcodec \ --output_dir=/home/riftuser/outputs_matrix/smolvla/task1-unknown_bs64_s20000 \ --job_name=smolvla_task1-unknown_bs64 \ --batch_size=64 \ --steps=20000 \ --log_freq=200 \ --save_freq=5000 \ --save_checkpoint=true \ --num_workers=8 \ --wandb.enable=true \ --wandb.project=so101-layout-generalization \ --wandb.mode=online \ --wandb.disable_artifact=true \ --policy.path=lerobot/smolvla_base \ --policy.device=cuda \ --policy.push_to_hub=false \ --rename_map={"observation.images.overhead": "observation.images.camera1", "observation.images.wrist": "observation.images.camera2"} ``` ## Repository Contents - `pretrained_model/`: final downloadable model artifacts for inference/loading - `training_state/`: optimizer, RNG, scheduler/state, and step information for resuming or auditability ## Notes - This repo stores the final checkpoint that was uploaded from the cloud training workspace. - The checkpoint was trained with LeRobot tooling via `lerobot-train`. - For SO101 experiments in this workspace, the dataset source was created by Aswinkumar. ## Creator Aswinkumar - Website: [aswinkumar.me](https://aswinkumar.me) - Hugging Face repo: