Registry / ai-ml / nvidia-modelopt-core

nvidia-modelopt-core

library0.33.1pypypiunverified

The NVIDIA TensorRT Model Optimizer (ModelOpt) provides a unified toolkit for model optimization and deployment across NVIDIA GPUs, supporting quantization (PTQ, QAT), pruning, distillation, and TensorRT export. As of v0.33.1, the library is actively maintained and targets Python 3.10–3.12. Release cadence is approximately monthly.

ai-ml

pip install nvidia-modelopt-core

github ↗homepage ↗

INSTALL

IMPORT

SIG · NVIDIA-MODELOPT-CO

nvidia-modelopt-core

ai-mlpythonv0.33.1

harness data pending

Install & Compatibility

Where this runs

No compatibility data collected yet for this library.

Code

Verified usage

Verified import paths — ran on the pinned version, not inferred.

ModelOptTorchModel

✓ from modelopt_core.torch.opt import ModelOptTorchModel

✗ from modelopt.torch.opt import ModelOptTorchModel

Basic scaffold to apply INT8 quantization to a PyTorch model. Replace with actual calibration data.

import torch import modelopt.torch as mt # Load a pretrained model (e.g., resnet50) model = torch.hub.load('pytorch/vision:v0.10.0', 'resnet50', pretrained=True) # Apply INT8 PTQ (requires calibration data) # For demonstration, we skip calibration and just show the pattern # quantized_model = mt.quantization.quantize(model, mode='int8', calibrator='max') print('Model optimization scaffold ready')

Debug

Known issues

breakingThe library was previously distributed as `nvidia-modelopt` (v0.x) and has been renamed to `nvidia-modelopt-core` starting with v0.33.0. The old package name is no longer updated and imports from `nvidia_modelopt` will fail.

fix

Uninstall `nvidia-modelopt` and install `nvidia-modelopt-core`. Update imports to use `modelopt.*` instead of `nvidia_modelopt.*`.

affects: >=0.33.0

gotchaThe `modelopt.torch.quantization.quantize` function returns a new model object; the original model is not modified in-place.

fix

Assign the return value: `quantized_model = quantize(model, ...)`.

affects: All

deprecatedThe `modelopt.torch.pruning` module is deprecated in favor of `modelopt.torch.sparsity` for pruning workflows as of v0.30.0.

fix

Use `modelopt.torch.sparsity` instead of `modelopt.torch.pruning`.

affects: >=0.30.0

gotchaTensorRT export (`modelopt.onnx.export`) requires ONNX opset >=17 and may fail with older opsets. Ensure your model is exported with opset 17 or later.

fix

When exporting to ONNX, set `opset_version=17` in `torch.onnx.export`.

affects: All

Upgrade

Version history

0.33.1latest on PyPI · released Aug 12, 2025

Audit

Dependencies

torchoptionalRequired for PyTorch model optimization workflows

onnxoptionalRequired for ONNX model support

tensorrtoptionalRequired for TensorRT engine building and inference

Agent activity

6 hits · last 30 days

node

Resources

githubhttps://github.com/NVIDIA/TensorRT-Model-Optimizer ↗

homepagehttps://github.com/NVIDIA/TensorRT-Model-Optimizer ↗