Quantize pretrained models to nvfp4, fp8, or int4_awq checkpoints using ModelOpt PTQ.
npx skills add https://github.com/NVIDIA/skills --skill ptq
Please help me install this Agent Skill. Skill: ptq Source: https://github.com/NVIDIA/skills/tree/main/skills/Model-Optimizer/ptq Command: npx skills add https://github.com/NVIDIA/skills --skill ptq
Post-training quantization (PTQ) enables converting pretrained models into compact, deployment-friendly checkpoints with reduced memory footprint and faster inference using ModelOpt.
Quantize a pretrained model to produce a deployment-ready quantized checkpoint using ModelOpt PTQ.
High-intent search queries and answers about installing and using this skill.
To quantize a pretrained model with ModelOpt, you run the PTQ workflow which automates calibration and validation to export a deployment-ready checkpoint in formats like fp8 or nvfp4.
Post-training quantization (PTQ) converts pretrained models into compact, deployment-friendly checkpoints, reducing memory footprint and accelerating inference without requiring retraining.
Yes, ModelOpt PTQ supports quantizing HuggingFace, TensorRT-LLM backbones, and both MoE and VLM architectures to formats such as nvfp4, fp8, and int4_awq.
You must prepare the environment, verify model support, execute optional remote or launcher runs for calibration, and validate the checkpoint against the model to ensure the correct quantization pattern.
Post-training quantization requires validation to ensure the exported checkpoint applies the correct quantization pattern and maintains deployment-ready accuracy after calibration.
yo-stevenQuantize pretrained LLMs and VLMs into HuggingFace or TensorRT-LLM checkpoints.
nadicodeaiDeploy high-throughput LLM inference on NVIDIA GPUs with tensorrt-llm.
NITISH-gitbitOptimize LLM inference on NVIDIA GPUs with TensorRT and FP8/INT4 quantization.
QingHuan-6Quantize diffusion DiT models with ModelOpt to FP8 or NVFP4 checkpoints for SGLang Diffusion.
