Train and evaluate Neural Frame Rate Upscaling models using Model Gym
Introduction
Understand neural graphics and Model Gym
Set up your environment
Launch the training notebook
Fine-tune and export a quantized model
Visualize your model with Model Explorer
Next Steps

  Log an issue

  Fork and edit

  Discuss on Discord

About this Learning Path

Skill level:  Advanced
Reading time:  45 min
Last updated:  24 Aug 2026
Skill level:
  Advanced
Reading time:
  45 min
Last updated:
  24 Aug 2026
Author:Annie Tallund, Arm
Arm IP:Mali
Tags: ML Linux PyTorch Jupyter Notebook Vulkan NX
Author:
  • Annie Tallund, Arm
Arm IP:
Mali
Tags:
ML Linux PyTorch Jupyter Notebook Vulkan NX

Who is this for?

This is an advanced topic for developers exploring neural graphics and interested in training and deploying frame generation models such as Neural Frame Rate Upscaling (NFRU) using PyTorch and Arm's hardware-aware backend.

What will you learn?

Upon completion of this Learning Path, you will be able to:

  • Understand the principles of neural graphics and how it's applied to game performance
  • Fine-tune and evaluate a neural network for Neural Frame Rate Upscaling (NFRU)
  • Use the Model Gym Python API and CLI to configure and train neural graphics models
  • Fine-tune an NFRU model with quantization-aware training (QAT) and export it to .vgf
  • Inspect the graph of exported .vgf models using Model Explorer

Prerequisites

Before starting, you will need the following:

  • Basic understanding of PyTorch and machine learning concepts
  • A development machine running Ubuntu 22.04, with a CUDA-capable NVIDIA GPU
  • CUDA Toolkit version 11.8 or later
  • A working environment with a Python version later than 3.10

Summary

AI-assisted

This summary was drafted with an approved AI-assisted workflow and reviewed by Arm contributors before publication. Human technical review remains part of the process so the final page reflects engineering rigor, accuracy, and Arm editorial standards.

Close
?
You’ll explore neural graphics and use Arm Neural Graphics Model Gym to train and evaluate an NFRU model. First, you’ll set up Ubuntu, clone the examples repository, and use Jupyter notebooks to train, evaluate, and compare checkpoints. Then, you’ll fine-tune pretrained FP32 weights with QAT and export an INT8 .vgf model. Finally, you’ll use Model Explorer and the VGF adapter to inspect the model’s architecture, tensor shapes, and graph connectivity.

Frequently asked questions

AI-assisted

These FAQs were drafted with an approved AI-assisted workflow and reviewed by Arm contributors before publication. Human technical review remains part of the process so the final page reflects engineering rigor, accuracy, and Arm editorial standards.

Close
?
What should I check before I run the notebooks on Ubuntu?
Use Ubuntu 22.04 with a CUDA-capable NVIDIA GPU, CUDA Toolkit 11.8 or later, and Python later than 3.10. Then, install the required dependency packages.
Where do I start the training after cloning the examples repository?
From neural-graphics-model-gym-examples, run jupyter lab. Open tutorials/nfru/nfru_training_example.ipynb and step through the notebook for training.
How do I know the initial training produced usable artifacts?
Create and inspect PyTorch checkpoints in the training and evaluation notebooks. In the evaluation notebook, you can measure accuracy, compare checkpoints, and see a visual comparison of the generated NFRU frame with the ground truth frame.
Should I use QAT or Post-training quantization (PTQ) for deployment?
Start with PTQ if you want a faster trial because it calibrates an already-trained model without another training phase. Use QAT when PTQ causes unacceptable accuracy or visual-quality regressions. QAT simulates lower-precision inference during fine-tuning and helps preserve accuracy when you quantize the model to INT8.
How do I export and inspect the deployable model?
Use the QAT notebook from the dedicated NFRU examples tag. The notebook includes the export step to produce a .vgf file. Then, open the .vgf file in Model Explorer with the VGF adapter to inspect architecture, tensor shapes, and graph connectivity.
Next