README.md

---
license: mit
datasets:
- ylecun/mnist
tags:
- harley-ml
- image
- digit-to-image
- mnist
- small
- text-to-image
- supralabs
---

# **SupraMNiST-IMG-200k**

## Sumary

```
Task: Number-To-Image
Dataset: ylecun/mnist
Total training time: ~8 minutes
Inputs: Number (0-9) 
Outputs: 32x32 image
Params: ~201k
Framework: PyTorch, diffusers
Author: SupraLabs
```

## **Description**
MNiST-IMG-200k is an ~**200k parameter model** trained to **generate an image** based on an **input number (0-9)**. 

## Architecture

| Parameter            | Value      |
| -------------------- | ---------- |
| `image_size`         | `32`       |
| `in_channels`        | `1`        |
| `out_channels`       | `1`        |
| `num_classes`        | `10`       |
| `block_out_channels` | `[12, 16]` |
| `layers_per_block`   | `8`        |
| `norm_num_groups`    | `4`        |

## **Training**

### **Hardware**

MNiST-IMG was trained on Google Colaboratory (NVIDA Tesla T4) for ~8 minutes with a batch size of 64 for 10 epochs.

### **Dataset**

[ylecun/mnist](https://huggingface.co/ylecun/mnist)

### **Training Results**

Loss ended at ~**0.40**.

Note: I can't provide the raw training logs as I loss it somehwere after training. Sorry!

## **Generation Examples**

At **1000** decoding steps:

![1000 Decoding Step Digit Image Generation](images/digit_image_samples_1000s.png)

At **200** decoding steps:

![200 Decoding Step Generation Image](images/digit_image_samples_200s.png)

# Inference

Use the script in the repo. [inference.py](https://huggingface.co/Harley-ml/MNIST-IMG-390k/blob/main/inference.py)

### Related Models

1. [MNIST-IMG-390k](https://huggingface.co/Harley-ml/MNIST-IMG-390k)

# Citation

```bibtex
@misc{mnist-img-390k,
  title     = {MNIST-IMG-390k: a Tiny Diffusion Model for Generating Handwritten Digits},
  author    = {Paul Courneya; Harley-ml},
  year      = {2026},
  url       = {https://huggingface.co/Harley-ml/MNIST-IMG-390k}
}
```