SmolVLA finetuned on SO-ARM100 dump-pocket-cleaning
This is a SmolVLA policy finetuned with LeRobot.
Training summary
- policy:
smolvla - pretrained base:
lerobot/smolvla_base - dataset:
Autobrik/SO-ARM100-dump-pocket-cleaning2 - run_dir:
outputs/train/benchmark/20260215_173100_benchmark_smolvla - seed:
1000
Best offline validation
- best_val_step:
14728 - best_val_loss:
0.025387765275310493
Included artifacts
checkpoints/best(recommended for eval/deploy)checkpoints/last(resume/debug)best_val.json