act_carton_phase_c_50k

ACT policy, 50,000 steps, from the Phase-C augmented holdout run act_phase_c_aug_holdout_20260824_150345. Carries its own train_config.json, so the dataset and hyperparameters it was trained with are recoverable from the checkpoint itself.

Published because it is the base that act_carton_middle_labels_5k was fine-tuned from — without it that fine-tune cannot be reproduced or explained. The 80k checkpoint from the same family is act_carton_phase_c_80k.

Code, deployment wrappers and gate status: steven-tired/mediapipe-so101.

Downloads last month
4
Safetensors
Model size
51.6M params
Tensor type
F32
·
Video Preview
loading