README.md

---

pipeline_tag: image-classification
license: apache-2.0
base_model: timm/deit_tiny_patch16_224.fb_in1k
library_name: zeromodels
tags:
- keras
- zeromodels
- image-classification
- deit
- backbone
- arxiv:2012.12877
- pytorch
- jax
- tf
---


## ***See [our collection](https://huggingface.co/collections/zeromodels/deit-and-deit3-6a8eaefd844d1822f01d81b2) for all versions of DeiT.***

# Run DeiT with Keras 3: JAX, PyTorch, or TensorFlow

[![GitHub](https://img.shields.io/badge/GitHub-ZeroModels-black?logo=github)](https://github.com/IMvision12/ZeroModels) [![Docs](https://img.shields.io/badge/Docs-DeiT-blue)](https://imvision12.github.io/ZeroModels/classification_backbones/) [![Collection](https://img.shields.io/badge/HF-DeiT%20collection-yellow)](https://huggingface.co/collections/zeromodels/deit-and-deit3-6a8eaefd844d1822f01d81b2)

# zeromodels/deit_tiny_patch16_224_fb_in1k



Paper: [Training data-efficient image transformers and distillation through attention (arXiv:2012.12877)](https://arxiv.org/abs/2012.12877) · [HF Papers](https://huggingface.co/papers/2012.12877)



DeiT / DeiT3 are data-efficient ViT variants (distillation token in DeiT; improved training recipe in DeiT3). Same ImageClassify / Model API.



For more details on the model, please go to the upstream [model card](https://huggingface.co/timm/deit_tiny_patch16_224.fb_in1k).



Pure-**Keras 3** conversion of [`timm/deit_tiny_patch16_224.fb_in1k`](https://huggingface.co/timm/deit_tiny_patch16_224.fb_in1k) for [zeromodels](https://github.com/IMvision12/ZeroModels). One implementation runs unmodified on **TensorFlow / Torch / JAX**.



This is an **image-classification / backbone** checkpoint (`DeiTImageClassify` / `DeiTModel`).



## ✨ Quick start



```python

import os



os.environ["KERAS_BACKEND"] = "torch"  # or "jax" / "tensorflow"

from PIL import Image
from zeromodels.models.deit import DeiTImageClassify, DeiTModel, DeiTImageProcessor

model = DeiTImageClassify.from_weights("zeromodels/deit_tiny_patch16_224_fb_in1k")
processor = DeiTImageProcessor.from_weights("zeromodels/deit_tiny_patch16_224_fb_in1k")

image = Image.open("your_image.jpg").convert("RGB")

pixels = processor(image)  # resize + normalize (normalization lives in the processor)

logits = model(pixels, training=False)

print(logits.shape)  # (1, num_classes)

# Feature extraction: the backbone without the classifier head
backbone = DeiTModel.from_weights("zeromodels/deit_tiny_patch16_224_fb_in1k", as_backbone=True)

features = backbone(pixels, training=False)

```



Load any DeiT variant the same way with `from_weights("zeromodels/<variant>")`:

| Variant | Hub |
|---|---|
| `deit3_base_patch16_224_fb_in1k` | [`zeromodels/deit3_base_patch16_224_fb_in1k`](https://huggingface.co/zeromodels/deit3_base_patch16_224_fb_in1k) |
| `deit3_base_patch16_224_fb_in22k_ft_in1k` | [`zeromodels/deit3_base_patch16_224_fb_in22k_ft_in1k`](https://huggingface.co/zeromodels/deit3_base_patch16_224_fb_in22k_ft_in1k) |
| `deit3_base_patch16_384_fb_in1k` | [`zeromodels/deit3_base_patch16_384_fb_in1k`](https://huggingface.co/zeromodels/deit3_base_patch16_384_fb_in1k) |
| `deit3_base_patch16_384_fb_in22k_ft_in1k` | [`zeromodels/deit3_base_patch16_384_fb_in22k_ft_in1k`](https://huggingface.co/zeromodels/deit3_base_patch16_384_fb_in22k_ft_in1k) |
| `deit3_huge_patch14_224_fb_in1k` | [`zeromodels/deit3_huge_patch14_224_fb_in1k`](https://huggingface.co/zeromodels/deit3_huge_patch14_224_fb_in1k) |
| `deit3_huge_patch14_224_fb_in22k_ft_in1k` | [`zeromodels/deit3_huge_patch14_224_fb_in22k_ft_in1k`](https://huggingface.co/zeromodels/deit3_huge_patch14_224_fb_in22k_ft_in1k) |
| `deit3_large_patch16_224_fb_in1k` | [`zeromodels/deit3_large_patch16_224_fb_in1k`](https://huggingface.co/zeromodels/deit3_large_patch16_224_fb_in1k) |
| `deit3_large_patch16_224_fb_in22k_ft_in1k` | [`zeromodels/deit3_large_patch16_224_fb_in22k_ft_in1k`](https://huggingface.co/zeromodels/deit3_large_patch16_224_fb_in22k_ft_in1k) |
| `deit3_large_patch16_384_fb_in1k` | [`zeromodels/deit3_large_patch16_384_fb_in1k`](https://huggingface.co/zeromodels/deit3_large_patch16_384_fb_in1k) |
| `deit3_large_patch16_384_fb_in22k_ft_in1k` | [`zeromodels/deit3_large_patch16_384_fb_in22k_ft_in1k`](https://huggingface.co/zeromodels/deit3_large_patch16_384_fb_in22k_ft_in1k) |
| `deit3_medium_patch16_224_fb_in1k` | [`zeromodels/deit3_medium_patch16_224_fb_in1k`](https://huggingface.co/zeromodels/deit3_medium_patch16_224_fb_in1k) |
| `deit3_medium_patch16_224_fb_in22k_ft_in1k` | [`zeromodels/deit3_medium_patch16_224_fb_in22k_ft_in1k`](https://huggingface.co/zeromodels/deit3_medium_patch16_224_fb_in22k_ft_in1k) |
| `deit3_small_patch16_224_fb_in1k` | [`zeromodels/deit3_small_patch16_224_fb_in1k`](https://huggingface.co/zeromodels/deit3_small_patch16_224_fb_in1k) |
| `deit3_small_patch16_224_fb_in22k_ft_in1k` | [`zeromodels/deit3_small_patch16_224_fb_in22k_ft_in1k`](https://huggingface.co/zeromodels/deit3_small_patch16_224_fb_in22k_ft_in1k) |
| `deit3_small_patch16_384_fb_in1k` | [`zeromodels/deit3_small_patch16_384_fb_in1k`](https://huggingface.co/zeromodels/deit3_small_patch16_384_fb_in1k) |
| `deit3_small_patch16_384_fb_in22k_ft_in1k` | [`zeromodels/deit3_small_patch16_384_fb_in22k_ft_in1k`](https://huggingface.co/zeromodels/deit3_small_patch16_384_fb_in22k_ft_in1k) |
| `deit_base_distilled_patch16_224_fb_in1k` | [`zeromodels/deit_base_distilled_patch16_224_fb_in1k`](https://huggingface.co/zeromodels/deit_base_distilled_patch16_224_fb_in1k) |
| `deit_base_distilled_patch16_384_fb_in1k` | [`zeromodels/deit_base_distilled_patch16_384_fb_in1k`](https://huggingface.co/zeromodels/deit_base_distilled_patch16_384_fb_in1k) |
| `deit_base_patch16_224_fb_in1k` | [`zeromodels/deit_base_patch16_224_fb_in1k`](https://huggingface.co/zeromodels/deit_base_patch16_224_fb_in1k) |
| `deit_base_patch16_384_fb_in1k` | [`zeromodels/deit_base_patch16_384_fb_in1k`](https://huggingface.co/zeromodels/deit_base_patch16_384_fb_in1k) |
| `deit_small_distilled_patch16_224_fb_in1k` | [`zeromodels/deit_small_distilled_patch16_224_fb_in1k`](https://huggingface.co/zeromodels/deit_small_distilled_patch16_224_fb_in1k) |
| `deit_small_patch16_224_fb_in1k` | [`zeromodels/deit_small_patch16_224_fb_in1k`](https://huggingface.co/zeromodels/deit_small_patch16_224_fb_in1k) |
| `deit_tiny_distilled_patch16_224_fb_in1k` | [`zeromodels/deit_tiny_distilled_patch16_224_fb_in1k`](https://huggingface.co/zeromodels/deit_tiny_distilled_patch16_224_fb_in1k) |
| `deit_tiny_patch16_224_fb_in1k` | [`zeromodels/deit_tiny_patch16_224_fb_in1k`](https://huggingface.co/zeromodels/deit_tiny_patch16_224_fb_in1k) |

## Tips

- Set `KERAS_BACKEND` **before** importing Keras / zeromodels.
- `DeiTImageClassify` returns class logits; `DeiTModel` returns features (`as_backbone=True` for multi-scale stages).
- See [docs](https://imvision12.github.io/ZeroModels/classification_backbones/) and [Loading Weights](https://imvision12.github.io/ZeroModels/loading_weights/).
- Upstream / timm checkpoints: `DeiTImageClassify.from_weights("hf:timm/deit_tiny_patch16_224.fb_in1k")`.

## Special Thanks

A huge thank you to the DeiT authors and the timm / Hub communities for creating and releasing these models.

License: see YAML `license` (usually matches the upstream checkpoint).