ptq

Use when the user asks to "quantize a model", "run PTQ", "post-training quantization", "NVFP4 quantization", "FP8 quantization", "INT8 quantization", "INT4 AWQ", "quantize LLM", "quantize MoE", "quantize VLM", or needs to produce a quantized HuggingFace checkpoint from a pretrained model using ModelOpt. Do NOT use for multi-candidate recipe exploration or optimization (use quant-recipe-search).

Quality

87%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Critical

Do not install without reviewing

Quality

Content

85%Weight 40%Scale 1-3

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

An actionable, well-sequenced PTQ workflow with strong validation checkpoints and clean progressive disclosure into verified reference files. The main weakness is some redundant restatement of the same caveats across multiple sections.

Suggestions

Consolidate the VLM vision-tower exclusion guidance into one location (Step 3 or Key API Rules) and cross-reference it rather than restating it in both.

Merge the Step 2.5 model-specific-dependency handling with the matching Common Pitfalls entry to avoid duplicate coverage of EXTRA_PIP_DEPS / manual install.

Trim the References table to files not already cited inline, or replace inline citations with pointers to the table, to remove the overlap.

Dimension	Reasoning	Score
Conciseness	The body is dense and high-signal with no basic-concept padding, but it repeats material across sections (the VLM vision-tower exclusion appears in Step 3 and Key API Rules; dependency handling in Step 2.5 and Common Pitfalls; the References table largely duplicates inline mentions) and could be tightened.	2 / 3
Actionability	It provides fully executable, copy-paste-ready commands with concrete flags (pip install, hf_ptq.py with --calib_size 512/--qformat, launcher env-var invocations, grep/ls checks), not pseudocode.	3 / 3
Workflow Clarity	It lays out a clear Step 1-5 sequence with sub-paths 4A/4B/4C and explicit validation checkpoints (recipe match-count sanity check, smoke test before full calibration, the required Step 5 validation gate) plus feedback loops (launcher failure falls back to 4A).	3 / 3
Progressive Disclosure	SKILL.md is a well-organized overview pointing to real, one-level-deep reference files (all verified present: checkpoint-validation.md, launcher-guide.md, slurm-setup-ptq.md, unsupported-models.md), each clearly signaled inline and indexed in a References table mapping file to when-to-read.	3 / 3
	Total	11 / 12 Passed

Description

90%Weight 40%Scale 1-3

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 description with explicit Use-when triggers, natural keywords, and clear disambiguation from a sibling skill. Its only weakness is that it states a single concrete action rather than enumerating multiple capabilities.

Dimension	Reasoning	Score
Specificity	It names the domain and one concrete action ("produce a quantized HuggingFace checkpoint from a pretrained model using ModelOpt") but does not list multiple distinct actions, so it is not comprehensive enough for a 3.	2 / 3
Completeness	It explicitly answers both what ("produce a quantized HuggingFace checkpoint...") and when ("Use when the user asks to...") with explicit trigger guidance.	3 / 3
Trigger Term Quality	It enumerates natural phrases a user would actually say ("quantize a model", "run PTQ", "quantize LLM/MoE/VLM", "INT4 AWQ") with good coverage of common variations.	3 / 3
Distinctiveness Conflict Risk	It carves a clear niche (ModelOpt PTQ) and explicitly disambiguates the closest sibling ("Do NOT use for multi-candidate recipe exploration... use quant-recipe-search"), making a wrong-skill trigger unlikely.	3 / 3
	Total	11 / 12 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: NVIDIA/Model-Optimizer
Path: .agents/skills/ptq/SKILL.md
Commit: 33d05b0

Reviewed: 15 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.