skills.rest
amd
amdOFFICIAL
Explore all skills by this author →
WebsiteGitHub

quark-onnx-quant-plan

Build hardware-aware Quark ONNX quantization plans from model analysis and deployment intent.

154|27|Updated Jun 19, 2024
Search Tags:#calibration#quantization#onnx#deployment gating#quark#preset selection
One-click install
npx skills add https://github.com/amd/Quark --skill quark-onnx-quant-plan
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: quark-onnx-quant-plan
Source: https://github.com/amd/Quark/tree/main/.claude/skills-impl/l1-atomic/onnx/quark-onnx-quant-plan
Command: npx skills add https://github.com/amd/Quark --skill quark-onnx-quant-plan

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill helps you turn an ONNX model analysis and deployment intent into a concrete, hardware-aware quantization plan without guessing which preset, calibration method, or algorithm to use.

Core Features & Use Cases

  • Preset selection: Chooses among Quark ONNX presets such as XINT8, A8W8, A16W8, BF16, BFP16, and MX-family options based on model characteristics and target hardware.
  • Calibration and algorithm planning: Recommends calibration methods like MinMax or PowerOfTwo_MinMSE and algorithms such as CLE, AdaRound, AdaQuant, BiasCorrection, or AutoMixprecision.
  • Deployment-target gating: Enforces constraints for CPU, CUDA, ROCm, AMD NPU CNN, and AMD NPU Transformer targets, including required PoF2 calibration and custom-op availability checks.
  • Use case: If you have a model analysis report and need to decide whether to deploy a CNN to Ryzen AI NPU with XINT8 or keep it on GPU with A8W8, this Skill helps you choose the safest viable plan.

Quick Start

Ask for a Quark ONNX quantization plan using your model_analysis.json, target hardware, and accuracy goals.

Frequently Asked Questions about quark-onnx-quant-plan

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I choose the right Quark ONNX quantization preset for my target hardware?▼

Quark ONNX quantization preset selection matches model characteristics and target hardware to options like XINT8, A8W8, A16W8, BF16, BFP16, or MX-family configurations to ensure safe PTQ planning.

What is the best way to plan ONNX calibration and quantization algorithms for AMD NPU deployment?▼

Planning ONNX calibration for AMD NPU deployment involves recommending methods like MinMax or PowerOfTwo_MinMSE and algorithms such as CLE, AdaRound, AdaQuant, BiasCorrection, or AutoMixprecision based on your model analysis and accuracy goals.

How does deployment-target gating work for ONNX quantization on CPU, CUDA, and ROCm?▼

Deployment-target gating for ONNX quantization enforces hardware constraints for CPU, CUDA, and ROCm targets, including required Power-of-Two calibration checks and custom-op availability, ensuring the quant plan is viable for your specific runtime.

Can I deploy a CNN model to Ryzen AI NPU using XINT8 or should I keep it on GPU with A8W8?▼

You can evaluate deploying a CNN to Ryzen AI NPU with XINT8 versus keeping it on GPU with A8W8 by applying deployment-target gating to compare hardware constraints, calibration requirements, and accuracy goals for the safest viable plan.

What are the limitations when applying AMD NPU Transformer workflows with Quark ONNX quantization?▼

Limitations for AMD NPU Transformer workflows involve satisfying strict hardware constraints, including required PoF2 calibration and custom-op availability checks, which may restrict preset selection if your model relies on unsupported operations.

Related Skills

Explore More Software Engineering Skills→
amdamd
154

quark-onnx-router

Routes Quark ONNX user requests to the smallest correct downstream workflow.

Official
Advanced
amdamd
154

quark-onnx-model-intake

Extracts ONNX model metadata needed for Quark PTQ planning and compatibility assessment.

Official
Advanced
amdamd
154

quark-onnx-ptq-workflow

Guide end-to-end post-training quantization of ONNX models with AMD Quark.

Official
Advanced
amdamd
154

quark-torch-ptq

Quantize PyTorch transformer models into deployment artifacts using AMD Quark PTQ workflows.

Official
Advanced