amdOfficial·31 Agent Skills Included

Quark

Guided deep learning model quantization for PyTorch and ONNX

Guides end-to-end quantization of PyTorch and ONNX models using AMD Quark, from environment checks and installation to planning, execution, and validation. Removes guesswork around quantization schemes, calibration methods, hardware targets, and export formats like GGUF and safetensors. Diagnoses common failures such as CUDA OOM, missing execution providers, and version mismatches with concrete fixes.

154|27

↗

npx skills add amd/Quark --all -g -y

All Skills in This Repository (31)

Pure Emerald Level Indicators

📦 In Repo

amd

quark-onnx-autosearch-pro

Plans and orchestrates ONNX quantization auto-search runs with Optuna-driven presets and trial budgets.

quark-torch-llm-ptq-eval

Orchestrate post-training quantization workflows for Torch LLMs with validation and accuracy evaluation.

quark-onnx-doc-drift-check

Compare Quark ONNX skill guidance against upstream docs and source.

quark-onnx-skill-sync

Detect upstream AMD Quark ONNX drift in skill guidance and contracts.

quark-onnx-eval-runner

Verify Quark ONNX skill routing, planning, artifacts, and recovery checks.

quark-torch-eval-runner

Verify Quark skill routing, planning, artifact, and recovery outputs.

quark-torch-skill-sync

Detects upstream Quark changes that invalidate skill instructions or contracts.

quark-torch-doc-drift-check

Compare Quark skill guidance against upstream docs and source to detect outdated commands, flags, and options.

quark-skill-creator

Create and restructure Quark Agent Skills to meet format and governance contracts.

quark-workspace-validate

Validate workspace paths, model references, and output directories before Quark workflows.

quark-env-preflight

Collect OS, Python, GPU, and accelerator facts for Quark installation readiness.

quark-onnx-ptq-workflow

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

Official

Advanced

Frequently Asked Questions

FAQPage Schema

How to install Quark skills?▼

Run `npx skills add amd/Quark --all -g -y` in your terminal to install all skills in this suite globally.

How to quantize a Llama or Qwen model with FP8?▼

Ask your agent to quantize the model with FP8, and the quark-torch-ptq workflow walks you through model intake, plan confirmation, and execution with checkpoints.

Can Quark quantize ONNX models for AMD NPU?▼

Yes. The quark-onnx-ptq skill handles ONNX-to-ONNX quantization with XINT8 presets and power-of-two calibration required for Ryzen AI NPU deployment.

How to check if quantization hurt model accuracy?▼

The quark-torch-llm-ptq-eval recipe validates the quantized output and runs accuracy benchmarks like gsm8k or perplexity, comparing scores against reference baselines.

Do these skills work without deep learning expertise?▼

Yes. The skills present decision tables with sensible defaults and require your confirmation before running anything, so you only need to state your goal in plain English.