YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

pi05 tomato full comparison

Cohort Selected / admitted source episodes Frames Complete h10 windows Updates
new_passed56 56 / 55 20857 17590 30000
new56_old47 103 / 102 39616 35179 30000

Revision ef73fd6b1ab54dadd9a9703b286f1f18c2ca2296, align_v2. One camera-timestamp-invalid episode was rejected in both arms. Mixed includes 47 old unreviewed episodes. Separate train-only normalization per arm. Completed final checkpoints; not validation-selected best. No final trajectory-fit or real-robot evaluation. Each version has full inference params/assets, CPU relocation evidence and loading instructions. Shared offline tokenizer: runtime_assets/big_vision/paligemma_tokenizer.model. Gemma terms, including Section 3.2, apply. NOTICE; code license.

Mixed new and old data, action horizons

All listed checkpoints completed 30,000 full fine-tuning updates. Each horizon includes its own train-only normalization.

Horizon Final checkpoint Selected / contributing recordings Complete windows
h16 full/new56_plus_old47/h16/20261109-step030000 103 / 102 32221
h32 full/new56_plus_old47/h32/20261109-step030000 103 / 102 25765
h10 v2/h10 146 / 145 50583
h24 v2/h24 146 / 145 39846
h32 v2/h32 146 / 145 35056
h10 v3/h10 196 / 195 64901
h24 v3/h24 196 / 195 52305
h32 v3/h32 196 / 195 46658

New recordings passed review; old recordings were unreviewed. See each version for loading instructions. CPU policy loading and finite parameters are checked; no real-robot task-success evaluation is included.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support