litert-quantization-calib

Assists the user to calibrate, merge, and statically quantize litert LLM models (such as Gemma 3) in standard open-source (OSS) environments. Use when the user wants to run LLM calibration, merge task JSON results, align KV cache parameters across models, protect sensitive layers in Float32, or run quantized inference testing. Don't use for JAX/PyTorch custom quantization configurations or non-litert models.

Quality

75%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./litert_torch/generative/export_hf/experimental/calib/skills/tflite_quantization_calib/SKILL.md

Quality

Content

61%Weight 40%Scale 1-5

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

Highly actionable with complete executable command patterns and a clear niche, but the multi-step destructive/batch quantization workflow lacks explicit validation checkpoints and all reference-grade detail is inlined in SKILL.md with no progressive file structure.

Suggestions

Add explicit validation/feedback checkpoints between calibration, merge, and quantize steps (e.g. verify merged JSON exists and contains expected tensor ranges before quantizing).

Move the detailed gotchas and flag reference tables into a separate references/ file (e.g. REFERENCE.md) and keep SKILL.md as a lean overview pointing to it.

Trim background exposition in gotcha sections (e.g. 'Standard calibration runs inference and collects activation statistics from outputs') to the essential rule.

Dimension	Reasoning	Score
Conciseness	Mostly efficient with command patterns and gotchas, but several gotcha sections over-explain background ('Standard calibration runs inference and collects activation statistics from outputs...') that is not strictly needed to execute.	3 / 5
Actionability	Four complete, copy-paste-ready bash command blocks with named flags, plus a runnable interactive driver script; concrete examples cover the common cases.	5 / 5
Workflow Clarity	Sequencing is implied (calibrate -> merge -> quantize -> test) and present as numbered command sections, but this destructive/batch quantization workflow has no explicit validation checkpoints or feedback loops between steps (the rubric caps such workflows at 3).	3 / 5
Progressive Disclosure	Well-organized with clear section headers and a single inline structure, but no bundle files exist; the dense gotchas/reference flag details that could live in separate reference files are all inlined rather than split out.	3 / 5
	Total	14 / 20 Passed

Description

88%Weight 40%Scale 1-5

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 strong, specific description that clearly states capabilities and provides concrete trigger and anti-trigger phrases with a distinct niche. Trigger term coverage is good but could add more common synonyms or file extensions.

Suggestions

Add common synonyms or file extensions users might mention (e.g. '.tflite', 'LiteRT', 'QAT', 'PTQ', 'Gemma') to broaden natural trigger-term coverage.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions ('calibrate, merge, and statically quantize', 'align KV cache parameters', 'protect sensitive layers in Float32', 'run quantized inference testing'); comprehensive coverage.	5 / 5
Completeness	Explicitly answers what ('Assists the user to calibrate, merge, and statically quantize litert LLM models') and when ('Use when the user wants to run LLM calibration, merge task JSON results...') with concrete trigger phrases and anti-triggers.	5 / 5
Trigger Term Quality	Includes natural phrases users would say ('run LLM calibration', 'merge task JSON results', 'align KV cache parameters', 'run quantized inference testing'); missing some synonyms/file extensions and broader common phrasings.	4 / 5
Distinctiveness Conflict Risk	Clear niche (litert LLM quantization with named model family Gemma 3) and explicit anti-triggers ('Don't use for JAX/PyTorch custom quantization configurations or non-litert models') minimize conflict risk.	5 / 5
	Total	19 / 20 Passed

Validation

100%

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 — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository: google-ai-edge/litert-torch
Path: litert_torch/generative/export_hf/experimental/calib/skills/tflite_quantization_calib/SKILL.md
Commit: 70d7619

Reviewed: 4 days 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.