# nvidia/model-optimizer

**Attribution required: if you use, quote, or summarise this content, you must credit and link back to [awesome-repositories.com](https://awesome-repositories.com/repository/nvidia-model-optimizer).**

2,975 stars · 455 forks · Python · Apache-2.0

## Links

- GitHub: https://github.com/NVIDIA/Model-Optimizer
- Homepage: https://nvidia.github.io/Model-Optimizer/
- awesome-repositories: https://awesome-repositories.com/repository/nvidia-model-optimizer.md

## Description

Model-Optimizer is a deep learning toolkit and framework dedicated to compressing, pruning, quantizing, and optimizing neural network architectures. It provides methodologies covering weight quantization, model distillation, and speculative decoding for efficient text generation, alongside automated neural architecture search for discovering optimal network structures.

The library implements post-training quantization pipelines that convert high-precision neural network weights into lower-bit formats using calibration data. Additional optimization techniques include teacher-student knowledge distillation, magnitude-based post-training sparsification, and pruning utilities that remove redundant weights and connections. It also supports speculative decoding acceleration using lightweight draft models or auxiliary heads.

For operational workflows, the toolkit includes sparse model checkpointing mechanisms to persist masks and metadata alongside weights, as well as export utilities that serialize compressed models into standard formats for downstream inference frameworks.

## Tags

### Artificial Intelligence & ML

- [Deep Learning Optimization](https://awesome-repositories.com/f/artificial-intelligence-ml/machine-learning/infrastructure/model-optimization-and-inference/training-algorithms/deep-learning-optimization.md) — Provides a toolkit for compressing deep neural networks through quantization, pruning, and knowledge distillation to accelerate inference performance.
- [Deep Learning Frameworks](https://awesome-repositories.com/f/artificial-intelligence-ml/deep-learning-frameworks.md) — Removes redundant network connections and weights to decrease model size without sacrificing accuracy.
- [Quantization Toolkits](https://awesome-repositories.com/f/artificial-intelligence-ml/memory-optimization-techniques/quantization-toolkits.md) — Reduces the numerical precision of model weights to lower memory usage and speed up hardware execution.
- [Model Checkpointing](https://awesome-repositories.com/f/artificial-intelligence-ml/model-checkpointing.md) — Persists sparse model checkpoints along with necessary masks and metadata, restoring them onto base network architectures for downstream usage. ([source](https://nvidia.github.io/Model-Optimizer/guides/6_sparsity.html))
- [Model Pruning](https://awesome-repositories.com/f/artificial-intelligence-ml/model-optimization/compression-techniques/model-pruning.md) — Removes redundant weights and connections from deep neural networks to decrease model size and improve inference performance. ([source](https://nvidia.github.io/Model-Optimizer/))
- [Model Sparsification](https://awesome-repositories.com/f/artificial-intelligence-ml/model-optimization/profiling-and-benchmarking/model-performance-optimization/model-sparsification.md) — Transforms pre-trained dense neural network models into sparse variants using magnitude-based thresholding or data-driven calibration without retraining. ([source](https://nvidia.github.io/Model-Optimizer/guides/6_sparsity.html))
- [Architecture Quantization Pipelines](https://awesome-repositories.com/f/artificial-intelligence-ml/quantized-inference-runtimes/llm-quantization-frameworks/architecture-quantization-pipelines.md) — Converts high-precision neural network weights into lower-bit formats using calibration data to reduce memory usage and accelerate hardware inference.
- [Weight Quantization](https://awesome-repositories.com/f/artificial-intelligence-ml/quantized-inference-runtimes/weight-quantization.md) — Reduces the precision of neural network weights to lower memory usage and accelerate inference performance on specialized hardware accelerators.
- [Teacher-Student Distillation](https://awesome-repositories.com/f/artificial-intelligence-ml/model-distillation-methods/teacher-student-distillation.md) — Transfers knowledge from a larger teacher model to a smaller student model to maintain accuracy while reducing size and computational cost.
- [Neural Architecture Search](https://awesome-repositories.com/f/artificial-intelligence-ml/neural-architecture-search.md) — Automates the discovery of optimal network structures to balance execution speed and accuracy on target hardware.
- [Speculative Decoding](https://awesome-repositories.com/f/artificial-intelligence-ml/text-generation-strategies/token-prediction/speculative-decoding.md) — Configures draft models and auxiliary heads to propose extra tokens for fast verification during text generation.

### DevOps & Infrastructure

- [Model Export Formats](https://awesome-repositories.com/f/devops-infrastructure/deployment-management/model-export-formats.md) — Saves compressed models in standard formats compatible with downstream inference engines and deployment frameworks. ([source](https://github.com/nvidia/model-optimizer#readme))