Quantize PyTorch transformer models into deployment artifacts using AMD Quark PTQ workflows.
npx skills add https://github.com/amd/Quark --skill quark-torch-ptq
Please help me install this Agent Skill. Skill: quark-torch-ptq Source: https://github.com/amd/Quark/tree/main/.claude/skills/quark-torch-ptq Command: npx skills add https://github.com/amd/Quark --skill quark-torch-ptq
This Skill streamlines the full post-training quantization workflow for PyTorch and HuggingFace transformer models, helping you move from a safetensors model to a quantized output without manually stitching together planning, script generation, and execution steps.
Ask to quantize your HuggingFace safetensors model with AMD Quark PTQ and generate the full PyTorch workflow through the quantized output.
High-intent search queries and answers about installing and using this skill.
Quantize HuggingFace LLMs end to end by applying AMD Quark post-training quantization workflows to safetensors models, generating planning steps, scripts, and quantized artifacts without requiring ONNX inputs.
Apply FP8 or INT4 quantization to PyTorch transformer models by inspecting the model, generating a quantization workflow, and producing a deployment-ready quantized artifact for common model families like Llama, Qwen, or Mistral.
Generating a quantization workflow from safetensors format involves model inspection, quantization path planning, and script generation for AMD Quark PTQ, culminating in a quantized output artifact ready for deployment preparation.
AMD Quark post-training quantization supports PyTorch-specific quantization tasks directly on HuggingFace safetensors models without requiring ONNX inputs, streamlining the pipeline from planning to quantized artifact generation.
Use post-training quantization for PyTorch transformer models when preparing Llama, Qwen, or Mistral architectures for deployment, reducing model size and computational requirements through FP8 or INT4 quantization modes.
End-to-end PyTorch LLM quantization workflows stop at the quantized output artifact and do not include validation or evaluation steps, requiring separate downstream processes for performance verification and accuracy assessment.
amdGuide end-to-end post-training quantization of ONNX models with AMD Quark.
ovachieverQuantize LLMs to 8-bit or 4-bit precision using BitsAndBytes.
amdValidate Quark quantization outputs via file, tensor, and config checks.
amdBuild hardware-aware Quark ONNX quantization plans from model analysis and deployment intent.
