Quantization

ModelOpt quantization toolkit supports quantization for NVIDIA’s hardware and software stack. Currently ModelOpt supports quantization in PyTorch and ONNX frameworks.

ModelOpt is based on simulated quantization in the original precision to simulate, test and optimize for the best trade-off between the accuracy of the model and different low-precision formats. To achieve actual speedups and memory savings, the model with simulated quantization can be exported to deployment frameworks, like TensorRT or TensorRT-LLM. Please refer to the TensorRT-Model-Optimizer GitHub repository for more details and examples.

Below, you can find the documentation for the quantization toolkit in ModelOpt:

• Basic Concepts
• Best practices to choose the right quantization methods
• PyTorch Quantization
• ONNX Quantization (Beta)