Create export and evaluation plans for Quark quantized models.
npx skills add https://github.com/amd/Quark --skill quark-torch-export
Please help me install this Agent Skill. Skill: quark-torch-export Source: https://github.com/amd/Quark/tree/main/.claude/skills-impl/l1-atomic/torch/quark-torch-export Command: npx skills add https://github.com/amd/Quark --skill quark-torch-export
This Skill helps turn a completed or planned Quark post-training quantization run into a usable deployment export plan, so the model can be packaged for HuggingFace, ONNX Runtime, or GGUF-based inference without guessing the right settings.
Ask for a deployment-ready export plan for your Quark quantized model, including the target format, output directory, and any post-quantization evaluation you want enabled.
High-intent search queries and answers about installing and using this skill.
To export a quantized model, provide an existing quantization run manifest or plan to generate a deployment export plan. This process selects the correct target format, applies format-specific weight packing, and defines output paths for ONNX or GGUF inference.
Setting up post-quantization evaluation involves attaching perplexity, task-based, or ROUGE-style scoring parameters to your export plan. This updates the workflow manifest with evaluation fields, ensuring the SafeTensors export is validated before deployment.
Yes, Quark export supports packaging models for vLLM or llama.cpp by generating format-specific export plans. It configures the appropriate HuggingFace SafeTensors or GGUF weight formats and output directories required by these inference engines.
Yes, you need an existing quantization context such as a quant plan or run manifest to create an export plan. The Skill extends this existing workflow manifest with export and evaluation fields rather than creating conflicting artifacts from scratch.
Updating a workflow manifest requires feeding your existing quantization run manifest into the export planning process. The Skill then extends this manifest with GGUF weight packing configurations, output paths, and downstream evaluation parameters for seamless deployment handoff.
amdExecute end-to-end post-training quantization workflows for ONNX models with AMD Quark.
williamzujkowskiMitigate security vulnerabilities in ML model deployment workflows.
amdQuantize PyTorch transformer models into deployment artifacts using AMD Quark PTQ workflows.
amdOrchestrate AMD Quark post-training quantization workflows for PyTorch and ONNX models.
