Enhance model card for Neon with key metadata, paper, code, and usage

README.md

@@ -1,3 +1,101 @@
----
-license: mit
----
+---
+license: mit
+pipeline_tag: unconditional-image-generation
+---
+
+# Neon: Negative Extrapolation From Self-Training Improves Image Generation
+
+This repository contains the models presented in the paper [Neon: Negative Extrapolation From Self-Training Improves Image Generation](https://huggingface.co/papers/2510.03597).
+
+**Official GitHub Repository**: [https://github.com/SinaAlemohammad/Neon](https://github.com/SinaAlemohammad/Neon)
+
+## Introduction
+
+Scaling generative AI models is bottlenecked by the scarcity of high-quality training data. The ease of synthesizing from a generative model suggests using (unverified) synthetic data to augment a limited corpus of real data for the purpose of fine-tuning in the hope of improving performance. Unfortunately, however, the resulting positive feedback loop leads to model autophagy disorder (MAD, aka model collapse) that results in a rapid degradation in sample quality and/or diversity.
+
+**Neon** (for Negative Extrapolation frOm self-traiNing) is a new learning method that turns the degradation from self-training into a powerful signal for self-improvement. Given a base model, Neon first fine-tunes it on its own self-synthesized data but then, counterintuitively, reverses its gradient updates to extrapolate away from the degraded weights. This approach corrects a predictable anti-alignment between synthetic and real data population gradients, better aligning the model with the true data distribution.
+
+Neon is remarkably easy to implement via a simple post-hoc merge, requires no new real data, works effectively with as few as 1k synthetic samples, and typically uses less than 1% additional training compute. It demonstrates universality across a range of architectures (diffusion, flow matching, autoregressive, and inductive moment matching models) and datasets (ImageNet, CIFAR-10, and FFHQ). Notably, on ImageNet 256x256, Neon elevates the xAR-L model to a new state-of-the-art FID of 1.02 with only 0.36% additional training compute.
+
+## Method
+
+![Algorithm 1: Neon — Negative Extrapolation from Self‑Training](https://github.com/SinaAlemohammad/Neon/raw/main/assets/algorithm.png)
+
+**In one line:** sample with your usual inference to form a synthetic set $S$; briefly fine-tune the reference model on $S$ to get $\theta_s$; then **reverse** that update with a merge $\theta_{\text{neon}}=(1+w)\,\theta_r - w\,\theta_s$ (small $w>0$), which cancels mode-seeking drift and improves FID.
+
+## Benchmark Performance
+
+Neon consistently improves performance across various model types and datasets.
+
+| Model type    | Dataset          | Base model FID | Neon FID (paper) | Download model                                                                                          |
+| ------------- | ---------------- | -------------: | ---------------: | ------------------------------------------------------------------------------------------------------- |
+| xAR-L         | ImageNet-256     |           1.28 |         **1.02** | [Download](https://huggingface.co/sinaalemohammad/Neon/resolve/main/Neon_xARL_imagenet256.pth)          |
+| xAR-B         | ImageNet-256     |           1.72 |         **1.31** | [Download](https://huggingface.co/sinaalemohammad/Neon/resolve/main/Neon_xARB_imagenet256.pth)          |
+| VAR d16       | ImageNet-256     |           3.30 |         **2.01** | [Download](https://huggingface.co/sinaalemohammad/Neon/resolve/main/Neon_VARd16_imagenet256.pth)        |
+| VAR d36       | ImageNet-512     |           2.63 |         **1.70** | [Download](https://huggingface.co/sinaalemohammad/Neon/resolve/main/Neon_VARd36_imagenet512.pth)        |
+| EDM (cond.)   | CIFAR-10 (32×32) |           1.78 |         **1.38** | [Download](https://huggingface.co/sinaalemohammad/Neon/resolve/main/Neon_EDM_conditional_CIFAR10.pkl)   |
+| EDM (uncond.) | CIFAR-10 (32×32) |           1.98 |         **1.38** | [Download](https://huggingface.co/sinaalemohammad/Neon/resolve/main/Neon_EDM_unconditional_CIFAR10.pkl) |
+| EDM           | FFHQ-64×64       |           2.39 |         **1.12** | [Download](https://huggingface.co/sinaalemohammad/Neon/resolve/main/Neon_EDM_FFHQ.pkl)                  |
+| IMM           | ImageNet-256     |           1.99 |         **1.46** | [Download](https://huggingface.co/sinaalemohammad/Neon/resolve/main/Neon_imm_imagenet256.pkl)           |
+
+## Sample Usage (Evaluation and Generation)
+
+To get started with evaluating the Neon-enhanced models, follow the quickstart steps from the [official GitHub repository](https://github.com/SinaAlemohammad/Neon).
+
+### 1) Environment Setup
+
+First, set up the required environment:
+
+```bash
+# from repo root
+conda env create -f environment.yml
+conda activate neon
+```
+
+### 2) Download Pretrained Models & FID Stats
+
+Download the pretrained Neon models and necessary FID statistics:
+
+```bash
+bash download_models.sh
+```
+
+This populates `checkpoints/` and `fid_stats/`.
+**Pretrained Neon models can also be downloaded from Hugging Face:** [https://huggingface.co/sinaalemohammad/Neon](https://huggingface.co/sinaalemohammad/Neon)
+
+### 3) Download VAE for xAR Models
+
+For xAR models, download the pre-trained VAE:
+
+```bash
+hf download xwen99/mar-vae-kl16 --include kl16.ckpt --local-dir xAR/pretrained
+```
+
+### 4) Example: Evaluate xAR-L @ ImageNet-256
+
+After setting up the environment and downloading models, you can run the evaluation script. This command will generate images and calculate FID:
+
+```bash
+# All examples assume 8 GPUs; adjust `--nproc_per_node` / batch sizes as needed.
+PYTHONPATH=xAR torchrun --standalone --nproc_per_node=8 xAR/calculate_fid.py \
+  --model xar_large \
+  --model_ckpt checkpoints/Neon_xARL_imagenet256.pth \
+  --cfg 2.3 --vae_path xAR/pretrained/kl16.ckpt \
+  --num_images 50000 --batch_size 64 --flow_steps 40 --img_size 256 \
+  --fid_stats fid_stats/adm_in256_stats.npz
+```
+
+Similar commands for VAR, EDM, and IMM models can be found in the [GitHub Quickstart section](https://github.com/SinaAlemohammad/Neon#quickstart).
+
+## Citation
+
+If you find Neon useful, please consider citing the paper:
+
+```bibtex
+@article{neon2025,
+  title={Neon: Negative Extrapolation from Self-Training for Generative Models},
+  author={Alemohammad, Sina and collaborators},
+  journal={arXiv preprint},
+  year={2025}
+}
+```