# NNGPT: Rethinking AutoML with Large Language Models

Source: [NNGPT: Rethinking AutoML with Large Language Models](<https://www.sota2.com/research/paper/692689aad6b0dd334293051a>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-06-02T05:02:08.515Z

Authors: Roman Kochnev, Waleed Khalid, Tolgay Atinc Uzun, Xi Zhang, Yashkumar Sanjaybhai Dhameliya, Furui Qin, Chandini Vysyaraju, Raghuvir Duvvuri, Avi Goyal, Dmitry Ignatov, Radu Timofte

Published: 2025-11-25T14:10:44.000Z

## Abstract

Building self-improving AI systems remains a fundamental challenge in the AI domain. We present NNGPT, an open-source framework that turns a large language model (LLM) into a self-improving AutoML engine for neural network development, primarily for computer vision. Unlike previous frameworks, NNGPT extends the dataset of neural networks by generating new models, enabling continuous fine-tuning of LLMs based on closed-loop system of generation, assessment, and self-improvement. It integrates within one unified workflow five synergistic LLM-based pipelines: zero-shot architecture synthesis, hyperparameter optimization (HPO), code-aware accuracy/early-stop prediction, retrieval-augmented synthesis of scope-closed PyTorch blocks (NN-RAG), and reinforcement learning. Built on the LEMUR dataset as an audited corpus with reproducible metrics, NNGPT emits from a single prompt and validates network architecture, preprocessing code, and hyperparameters, executes them end-to-end, and learns from result. The PyTorch adapter makes NNGPT framework-agnostic, enabling strong performance: NN-RAG achieves 73% executability on 1,289 targets, 3-shot prompting boosts accuracy on common datasets, and hash-based deduplication saves hundreds of runs. One-shot prediction matches search-based AutoML, reducing the need for numerous trials. HPO on LEMUR achieves RMSE 0.60, outperforming Optuna (0.64), while the code-aware predictor reaches RMSE 0.14 with Pearson r=0.78. The system has already generated over 5K validated models, proving NNGPT as an autonomous AutoML engine. Upon acceptance, the code, prompts, and checkpoints will be released for public access to enable reproducibility and facilitate community usage.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Hyperparameter Recommendation on LEMUR (evaluation)](<https://www.sota2.com/research/sota/hyperparameter-recommendation-on-lemur-evaluation.md>): Hyperparameter Recommendation; LEMUR (evaluation); RMSE: 0.603.

- [Image Classification on MNIST (test) (Custom Accuracy Statistics)](<https://www.sota2.com/research/sota/image-classification-on-mnist-test-custom-accuracy-statistics.md>): Image Classification; MNIST (test); Average Accuracy: 98.76.

- [Image Classification on SVHN (test) (Accuracy Statistics)](<https://www.sota2.com/research/sota/image-classification-on-svhn-test-accuracy-statistics.md>): Image Classification; SVHN (test); Average Accuracy: 81.48.

- [Image Classification on CIFAR-100 (test) (Accuracy Profile Statistics)](<https://www.sota2.com/research/sota/image-classification-on-cifar-100-test-accuracy-profile-statistics.md>): Image Classification; CIFAR-100 (test); Average Accuracy: 1.63.

- [Hyperparameter Recommendation on 17 CV Models (evaluation)](<https://www.sota2.com/research/sota/hyperparameter-recommendation-on-17-cv-models-evaluation.md>): Hyperparameter Recommendation; 17 CV Models (evaluation); RMSE: 0.642.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2511.20333>)

[PDF](<https://arxiv.org/pdf/2511.20333.pdf>)
