fine-tuning-serving-openpi

Fine-tune and serve Physical Intelligence OpenPI models (pi0, pi0-fast, pi0.5) using JAX or PyTorch backends for robot policy inference across ALOHA, DROID, and LIBERO environments. Use when adapting pi0 models to custom datasets, converting JAX checkpoints to PyTorch, running policy inference servers, or debugging norm stats and GPU memory issues.

Quality

100%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Quality

Content

100%

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A well-structured, executable skill body with sequenced workflows, explicit validation checkpoints, and clean progressive disclosure into real reference files. It is dense yet concise and free of padding.

Dimension	Reasoning	Score
Conciseness	Lean and action-oriented throughout; it assumes Claude's competence and does not pad with explanations of concepts Claude already knows, while the compact comparison tables each earn their place.	3 / 3
Actionability	Provides fully executable bash and Python snippets — clone/sync/train/serve commands and websocket client examples — that are copy-paste ready with concrete flags and config names.	3 / 3
Workflow Clarity	Each workflow has a numbered, copyable progress checklist with explicit validation checkpoints (verify-install, test-client, 'compute norm stats must run before every training launch', and the compute→train→serve→validate loop invariant).	3 / 3
Progressive Disclosure	The body is an overview that defers detail to a clearly signaled, one-level-deep 'Advanced topics' section linking five real reference files (all verified present), with content appropriately split.	3 / 3
	Total	12 / 12 Passed

Description

100%

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A highly specific, third-person description that names concrete actions, model variants, and environments, and gives an explicit 'Use when' trigger clause. It is clearly distinguishable from adjacent VLA skills and contains no fluff.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions — 'Fine-tune and serve', 'converting JAX checkpoints to PyTorch', 'running policy inference servers', 'debugging norm stats and GPU memory issues' — rather than vague language.	3 / 3
Completeness	Explicitly answers both 'what' (fine-tune and serve OpenPI models across named environments) and 'when' via an explicit 'Use when adapting... converting... running... or debugging...' trigger clause.	3 / 3
Trigger Term Quality	Covers natural terms a user would actually say — 'fine-tune', 'serve', 'pi0/pi0-fast/pi0.5', 'ALOHA/DROID/LIBERO', 'JAX', 'PyTorch', 'norm stats', 'GPU memory' — including common variations.	3 / 3
Distinctiveness Conflict Risk	A clear VLA/OpenPI robot-policy niche with distinct model and environment triggers, written in third person, that is unlikely to fire for unrelated skills.	3 / 3
	Total	12 / 12 Passed

Validation

87%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 14 / 16 Passed

Validation for skill structure

Criteria	Description	Result
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning
referenced_paths_exist	Referenced path issues: 12 missing	Warning

	Total	14 / 16 Passed

Repository: Orchestra-Research/AI-Research-SKILLs
Path: 18-multimodal/openpi/SKILL.md
Commit: 773a529

Reviewed: about 12 hours ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.