privacy: scrub host chip/OS from model card

e7b4f39 verified 5 days ago

7.54 kB

license: gemma
base_model: lerobot/pi05_base
pipeline_tag: robotics
library_name: coreai
extra_gated_heading: Access this Gemma Model Derivative
extra_gated_prompt: >-
  These weights are a Model Derivative of Gemma. Access and use are governed by
  the Gemma Terms of Use (https://ai.google.dev/gemma/terms) and the Gemma
  Prohibited Use Policy (https://ai.google.dev/gemma/prohibited_use_policy). By
  requesting access you agree to those terms and the Section 3.2 use
  restrictions, and you agree to carry these terms forward on any further
  distribution.
extra_gated_fields:
  I agree to the Gemma Terms of Use and the Gemma Prohibited Use Policy: checkbox
  I agree to carry these terms forward on any redistribution: checkbox
tags:
  - coreai
  - core-ai
  - coreai-fabric
  - aimodel
  - coreml
  - apple
  - apple-silicon
  - on-device
  - robotics
  - vla
  - vision-language-action
  - action

Canonical: kevinqz/Pi05-Base-CoreAI — source of truth.

Pi0.5 base Gemma (fabric)

⚠️ Robot policy — needs a matching robot to actuate. This is lerobot/pi05_base converted to an Apple Core AI .aimodel. Its output is a normalized action chunk for generalist_multi_robot. Run it on any other robot, or with mismatched calibration / normalization stats, and it emits floats that look valid but actuate garbage — the most dangerous silent-failure mode. It is a base checkpoint: fine-tune on your robot's data before expecting useful behavior.

An Apple Core AI conversion of lerobot/pi05_base — a robot policy that maps images + proprioceptive state (+ a language instruction, when the model uses one) to a continuous action chunk (flow_matching sampler). Produced by coreai-fabric and indexed by coreai-catalog.

Model facts

Field	Value
Parameters	3.3B
Architecture	transformer
Capabilities	vision-language-action, robotics
Embodiment	generalist_multi_robot
Sampling	flow_matching (10-step)
Quantization / precision	none / float16
On-disk size	6.0 GB
Asset kind	split-export policy (encode, denoise_step) + norm_stats
assetVersion	2.0

Use it — this needs host code you supply

A policy is not a chat model: there is no stock high-level Swift runtime for it. The bundle is the split-export shape — an encode graph (run once per observation) + a denoise_step graph (the host drives it num_steps times) + norm_stats.json (un-normalization). You supply the host loop (the N-step sampler + un-normalization) in Swift. Recommended integration: keep LeRobot's Python RobotClient for the servos/cameras/calibration, and run inference on-device — see the io_contract in the catalog for the exact tensors.

pip install coreai-catalog && coreai-catalog install pi05-base

Requirements

Deployment: macOS 27.0+ / iOS 27.0+, Xcode 27+. The asset serializes with minimum_os v27, so the on-device Swift runtime requires macOS/iOS 27+. A Mac on macOS 26 can convert and inspect it but not run it on-device.
Apple Silicon. A matching robot to actuate (see the banner).

Verification (action parity)

Gate A (structure): passed — the bundle's layout + metadata were validated on real hardware; the graphs load.
Gate B — action_parity: 96.2% min chunk-cosine vs the fp16 reference over 16 frames of recorded episodes, 10-step, fixed-noise (measured on Apple Silicon).
Near-zero-action note (full disclosure): the 96.2% is the raw MIN over all frames, dominated by frame(s) where the reference action is ~zero (the robot barely moves) and chunk-cosine is mathematically ill-conditioned. The median chunk-cosine is 99.9% and the max ABSOLUTE per-dim action error is 0.0193 (normalized) — i.e. behaviorally faithful everywhere. This model class is gated on the absolute action error (behaviorally meaningful) since cosine is undefined for ~zero actions; every raw number is reported here and in parity-report.json. fabric never fakes a number.
This certifies the export is numerically faithful to the source policy — it does NOT certify real-world task success, embodiment transfer, or closed-loop stability. Closed-loop success-rate eval (LIBERO/ManiSkill): not_run (a separate future gate). Reproduce with coreai-fabric verify.

Provenance

Field	Value
Base model	lerobot/pi05_base @ `7de663972b7817d2c4cf2d84c821153dfea772e9`
Converted by	`models/pi05/export.py` (version not reported)
Recipe	pi05-base (recipe_source: fabric)
Precision / quantization	float16 / none
Conversion date	2026-07-05

Machine-readable, in this repo: parity-report.json (gate results) · reproduce-manifest.json · LICENSE (upstream terms).

License and attribution

Weights © Google LLC and the upstream authors, licensed gemma — see the bundled LICENSE. This artifact is a converted derivative of the base policy: its weights were converted to Apple Core AI format. The conversion itself is community work.

These weights are a Model Derivative of Gemma — the upstream pi0/pi0.5 policy embeds PaliGemma/Gemma parameters. Redistribution and use are governed by the Gemma Terms of Use (bundled as LICENSE) and restricted by the Gemma Prohibited Use Policy (§3.2, incorporated by reference). Access is gated: you must accept these terms before downloading, and must carry them forward — including the mandated NOTICE and the §3.2 restrictions — on any further distribution (see NOTICE). Gemma and PaliGemma are trademarks of Google LLC, used here only descriptively; this conversion is not affiliated with, produced by, or endorsed by Google.

The on-device Core AI ecosystem

coreai-fabric — the reproducible recipe → .aimodel pipeline that produced this asset.
coreai-catalog — the index of Core AI models across the community, with provenance and integration snippets.
apple/coreai-models — Apple's official exporters and runtimes.
LeRobot — the upstream robotics ecosystem.

Not affiliated with Apple

Community conversion. Not produced, hosted, or endorsed by Apple. Apple and Core AI are trademarks of Apple Inc., used here only to describe the target runtime/format. This is an independent community conversion.

kevinqz
/

Pi05-Base-CoreAI