pi05_sweep_v2/01_freeze/01_full
π₀.₅ (Pi05) — Vision-Language-Action model with open-world generalization, from Physical Intelligence. See the π₀.₅ blog post.
Trained with LeDream (a fork of LeRobot).
Training
- Base model:
lerobot/pi05_base - Dataset:
DreamMachines/actuator_unboxing_20h_345 - Step (this revision):
210000 - Batch size:
32 - Total steps target:
210000 - Optimizer:
adamw(lr5e-05)
Available revisions
Every training checkpoint is uploaded as its own git revision; the latest is also on main. Load a specific step with:
from lerobot.policies.factory import make_policy
from lerobot.configs.policies import PreTrainedConfig
cfg = PreTrainedConfig.from_pretrained("DreamMachines/actuator_unboxing_20h_345_fullft_bs256", revision="v002000")
policy = make_policy(cfg)
Available: v100000, v110000, v120000, v130000, v140000, v150000, v160000, v170000, v180000, v190000, v200000, v210000
Quick eval on a real DK1
make eval MODEL=DreamMachines/actuator_unboxing_20h_345_fullft_bs256
See docs/source/policy_pi05_README.md for π₀.₅-specific notes.