Diffusion Policy — Genesis HR v2 train19 (qpos-absolute, step 6,250)

This repository contains the exact inference-only checkpoint leaf used for LeRobot Diffusion Policy in the 19-task rebuttal evaluation run hf19_eval5_act080000_full95_fast_20260812T0236Z.

  • Architecture: LeRobot Diffusion Policy
  • Training dataset: zhouqh/hr_v2, the 19 tasks available for this training run
  • Action convention: qpos-absolute
  • Training run: hr_v2_train19_full_deadline_20260811T0934Z_4a435cf
  • Final checkpoint: step 006250
  • Evaluation source commit: b12c5190e55ec31468e9743871dfc3869117ea50 for the first 15 tasks; neutral-task compatibility retry used ee31b7eb74341303a4dd5985743f90efd1d892ec
  • Evaluation artifacts: https://huggingface.co/datasets/zimplex/genesis-hr-bench-rebuttal-full19-hrv2-eval-20260812

Only files from the checkpoint's pretrained_model directory are included; optimizer state and other training-state files are intentionally excluded. SHA256SUMS records the exact uploaded payload.

The linked evaluation artifact snapshot is preliminary: four place_food_in_skillet_assist seed-7 trajectories were invalid due to a simulator instability, and a scoped repair/retry was pending when the snapshot was frozen.

Downloads last month
3
Safetensors
Model size
0.3B params
Tensor type
F32
·
Video Preview
loading

Dataset used to train zimplex/diffusion-genesis-hr-v2-train19-qpos-abs-step006250