Downloads · 30 days
0
Travor278/pi05-piperx-attempt-v2-acp-20k-r2
pi05-piperx-attempt-v2-acp-20k-r2 is a robotics model from Travor278. Use it for the robotics task on the model card, and read the license before you ship it in a product. It is set up for lerobot. The card lists the license as apache-2.0.
This repository contains the complete step-20,000 continuation state for the attempt-aware ACP v2 PiperX copper-screw policy. The successful platform run is v2sam-ego2exo-attempt-v2-acp-20k-r2; r1 stopped at its first…
Downloads · 30 days
0
Access
Public
Updated Sep 8, 2026
Repo size
40.5 GB
Likes
0
Public
Click a slice to open those files.
.safetensors40.5 GB · 100%
From the Hugging Face model README
This repository contains the complete step-20,000 continuation state for the
attempt-aware ACP v2 PiperX copper-screw policy. The successful platform run is
v2sam-ego2exo-attempt-v2-acp-20k-r2; r1 stopped at its first checkpoint because
the HDD quota was exhausted, before being restarted from the identical baseline
with SSD checkpoint staging.
d85c7cd84060a924b6ef10d055491c500c5714a21ddbb49e1dbdc828b7a74147.20260906.72c0b87dd8eb6ade37fba4817ab6d4fa379c17cd (training code was frozen at parent commit 37de4c0b59ec152a2626f9c3d1652bc444018438).d8c2c0f0115ce3456dbfa38f5529276c0e4bbe16.Frozen test (46 attempts, 184 legal 50-step anchors) used the deployment prompt
below. Normalized chunk MAE was 0.032656 (attempt-bootstrap 95% CI
[0.030567, 0.034804]); physical first-step MAE was 0.731886; smoothness error
was 0.405761. Versus the matched Data-only model, normalized/physical chunk MAE
was statistically tied, first-step MAE was slightly lower but not significant,
and smoothness error improved by -0.020137 (95% CI [-0.021792, -0.018551]).
Inference prompt:
Insert the copper screw into the black sleeve.
Advantage: positive
pretrained_model/ is the deployment checkpoint. training_state/ contains the
optimizer, scheduler, RNG and global-step state required for exact continuation.
The provenance/ directory contains dataset, outcome, split, sampler, Value,
validation-selection, frozen-test, Git and checksum evidence. Offline metrics are
not robot success-rate estimates.