---
pipeline_tag: zero-shot-image-classification
license: apache-2.0
base_model: google/siglip-base-patch16-384
library_name: zeromodels
tags:
- keras
- zeromodels
- siglip
- zero-shot-image-classification
- vision
- arxiv:2303.15343
- pytorch
- jax
- tf
---

## ***See [our collection](https://huggingface.co/collections/zeromodels/siglip-6a8eaf41ca17669e8cf9976b) for all versions of SigLIP.***

# Run SigLIP 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-SigLIP-blue)](https://imvision12.github.io/ZeroModels/siglip/) [![Collection](https://img.shields.io/badge/HF-SigLIP%20collection-yellow)](https://huggingface.co/collections/zeromodels/siglip-6a8eaf41ca17669e8cf9976b)

# zeromodels/siglip_base_p16_384

Paper: [Sigmoid Loss for Language Image Pre-Training (arXiv:2303.15343)](https://arxiv.org/abs/2303.15343) · [HF Papers](https://huggingface.co/papers/2303.15343)

SigLIP is a vision + text dual encoder trained with a pairwise sigmoid loss instead of CLIP's softmax contrastive loss. Per-pair training scales to large batches and often improves zero-shot accuracy at the same model size.

For more details on the model, please go to the upstream [model card](https://huggingface.co/google/siglip-base-patch16-384).

Pure-**Keras 3** conversion of [`google/siglip-base-patch16-384`](https://huggingface.co/google/siglip-base-patch16-384) for [zeromodels](https://github.com/IMvision12/ZeroModels). One implementation runs unmodified on **TensorFlow / Torch / JAX**.

This is a **zero-shot image-text** checkpoint (`SigLIPZeroShotClassify`): pass image(s) and text prompts at inference time.

## ✨ Quick start

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

from zeromodels.models.siglip import (
    SigLIPProcessor,
    SigLIPZeroShotClassify,
)

processor = SigLIPProcessor.from_weights("zeromodels/siglip_base_p16_384")
model = SigLIPZeroShotClassify.from_weights("zeromodels/siglip_base_p16_384")

labels = [
    "a photo of a cat",
    "a photo of a dog",
    "a photo of a car",
    "a photo of a living room",
]
inputs = processor(text=labels, image_paths="your_image.jpg")
output = model(
    {
        "images": inputs["images"],
        "token_ids": inputs["input_ids"],
    }
)
print(output["image_logits"].shape)
```

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

| Variant | Hub |
|---|---|
| `siglip_base_p16_224` | [`zeromodels/siglip_base_p16_224`](https://huggingface.co/zeromodels/siglip_base_p16_224) |
| `siglip_base_p16_256` | [`zeromodels/siglip_base_p16_256`](https://huggingface.co/zeromodels/siglip_base_p16_256) |
| `siglip_base_p16_multilingual_256` | [`zeromodels/siglip_base_p16_multilingual_256`](https://huggingface.co/zeromodels/siglip_base_p16_multilingual_256) |
| `siglip_base_p16_384` | [`zeromodels/siglip_base_p16_384`](https://huggingface.co/zeromodels/siglip_base_p16_384) |
| `siglip_base_p16_512` | [`zeromodels/siglip_base_p16_512`](https://huggingface.co/zeromodels/siglip_base_p16_512) |
| `siglip_large_p16_256` | [`zeromodels/siglip_large_p16_256`](https://huggingface.co/zeromodels/siglip_large_p16_256) |
| `siglip_large_p16_384` | [`zeromodels/siglip_large_p16_384`](https://huggingface.co/zeromodels/siglip_large_p16_384) |
| `siglip_so400m_p14_224` | [`zeromodels/siglip_so400m_p14_224`](https://huggingface.co/zeromodels/siglip_so400m_p14_224) |
| `siglip_so400m_p14_384` | [`zeromodels/siglip_so400m_p14_384`](https://huggingface.co/zeromodels/siglip_so400m_p14_384) |

## Tips

- Set `KERAS_BACKEND` **before** importing Keras / zeromodels.
- Prefer `Processor.from_weights(...)` so image size and tokenizer match the variant.
- Map processor `input_ids` to model `token_ids`. No padding mask is required.
- See [SigLIP docs](https://imvision12.github.io/ZeroModels/siglip/) and [Loading Weights](https://imvision12.github.io/ZeroModels/loading_weights/).
- Community / upstream safetensors still work via the `hf:` prefix, e.g. `SigLIPZeroShotClassify.from_weights("hf:google/siglip-base-patch16-384")`.

## Special Thanks

A huge thank you to the Google SigLIP authors for creating and releasing these models.

License: Apache 2.0.