# NVIDIA Model Optimizer

> Apache-2.0 quantization/pruning/NAS/distill for HF → TensorRT-LLM/vLLM

**Category:** Python  
**Pricing:** open-source  
**URL:** https://nvidia.github.io/Model-Optimizer/  
**GitHub:** https://github.com/NVIDIA/Model-Optimizer  
**Added:** 2026-09-26

---

## The Hook

NVIDIA Model Optimizer (ModelOpt) is the Apache-2.0 Python library for quantization, pruning, NAS, distillation, and speculative decoding on HF/PyTorch/ONNX models toward TensorRT-LLM, vLLM, or SGLang.

## What's Good

Apache-2.0 (NVIDIA/Model-Optimizer). Free via PyPI `nvidia-modelopt`. Composable optimize → export path for Hugging Face / PyTorch / ONNX into SGLang, TensorRT-LLM, TensorRT, or vLLM. Active NVFP4/QAD tutorials and Nemotron checkpoint recipes. ★4527 on GitHub.

## The Catch

NVIDIA stack gravity (TensorRT, Megatron, CUDA) — not an agent product. Serious runs need GPU/CUDA reality; CPU toy paths will not match the blog numbers. Ecosystem pull toward NVIDIA deploy targets even though the license is open.

## Verdict

Apache-2.0 model-optimization library when you compress HF/PyTorch models for TensorRT-LLM/vLLM/SGLang — skip if you wanted an agent framework.