README.md

---
pipeline_tag: fill-mask
license: mit
base_model: microsoft/deberta-base
library_name: kerasformers
tags:
- keras
- kerasformers
- deberta
- deberta-v1
- fill-mask
- text-encoder
- arxiv:2006.03654
- arxiv:2111.09543
- pytorch
- jax
- tf
---

## ***See [our collection](https://huggingface.co/collections/kerasformers/deberta-v1-v2-v3-6a6e90bac01e412b478562f3) for all versions of DeBERTa (v1 / v2 / v3).***

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

[![GitHub](https://img.shields.io/badge/GitHub-KerasFormers-black?logo=github)](https://github.com/IMvision12/KerasFormers) [![Docs](https://img.shields.io/badge/Docs-DeBERTa-blue)](https://imvision12.github.io/KerasFormers/deberta/) [![Collection](https://img.shields.io/badge/HF-DeBERTa%20collection-yellow)](https://huggingface.co/collections/kerasformers/deberta-v1-v2-v3-6a6e90bac01e412b478562f3)

# kerasformers/deberta_base

Papers: [DeBERTa: Decoding-enhanced BERT with Disentangled Attention (arXiv:2006.03654)](https://arxiv.org/abs/2006.03654) · [DeBERTaV3 (arXiv:2111.09543)](https://arxiv.org/abs/2111.09543) · [HF Papers](https://huggingface.co/papers/2006.03654)

DeBERTa is Microsoft's disentangled-attention text encoder (content + relative position). v1 uses byte-level BPE; v2/v3 use SentencePiece. v3 adds ELECTRA-style pretraining with gradient-disentangled embedding sharing. Import from `deberta` / `deberta_v2` / `deberta_v3` to match the generation.

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

Pure-**Keras 3** conversion of [`microsoft/deberta-base`](https://huggingface.co/microsoft/deberta-base) for [kerasformers](https://github.com/IMvision12/KerasFormers). One implementation runs unmodified on **TensorFlow / Torch / JAX**.

This is a **fill-mask / encoder** checkpoint (`DebertaMaskedLM`, v1 base). Task heads (sequence/token classify, QA, …) load via `hf:` fine-tunes.

## ✨ Quick start (fill-mask)

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

from kerasformers.models.deberta import (
    DebertaMaskedLM,
    DebertaTokenizer,
)

mlm = DebertaMaskedLM.from_weights("kerasformers/deberta_base")
tokenizer = DebertaTokenizer.from_weights("kerasformers/deberta_base")

inputs = tokenizer("The capital of France is [MASK].")
logits = mlm(inputs)  # (1, L, vocab_size)
mask = int((inputs["input_ids"][0] == tokenizer.mask_token_id).argmax())
print(tokenizer.decode([int(logits[0, mask].argmax())]))
```

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

| Variant | Hub | Generation |
|---|---|---|
| `deberta_base` | [`kerasformers/deberta_base`](https://huggingface.co/kerasformers/deberta_base) | v1 |
| `deberta_large` | [`kerasformers/deberta_large`](https://huggingface.co/kerasformers/deberta_large) | v1 |
| `deberta_v2_xlarge` | [`kerasformers/deberta_v2_xlarge`](https://huggingface.co/kerasformers/deberta_v2_xlarge) | v2 |
| `deberta_v2_xxlarge` | [`kerasformers/deberta_v2_xxlarge`](https://huggingface.co/kerasformers/deberta_v2_xxlarge) | v2 |
| `deberta_v3_xsmall` | [`kerasformers/deberta_v3_xsmall`](https://huggingface.co/kerasformers/deberta_v3_xsmall) | v3 |
| `deberta_v3_small` | [`kerasformers/deberta_v3_small`](https://huggingface.co/kerasformers/deberta_v3_small) | v3 |
| `deberta_v3_base` | [`kerasformers/deberta_v3_base`](https://huggingface.co/kerasformers/deberta_v3_base) | v3 |
| `deberta_v3_large` | [`kerasformers/deberta_v3_large`](https://huggingface.co/kerasformers/deberta_v3_large) | v3 |

## Tips

- Set `KERAS_BACKEND` **before** importing Keras / kerasformers.
- Prefer `Tokenizer.from_weights(...)` so vocab and mask token match.
- Do not mix packages across generations (v1 ≠ v2 ≠ v3).
- See [DeBERTa docs](https://imvision12.github.io/KerasFormers/deberta/) and [Loading Weights](https://imvision12.github.io/KerasFormers/loading_weights/).
- Community / upstream safetensors still work via the `hf:` prefix, e.g. `DebertaMaskedLM.from_weights("hf:microsoft/deberta-base")`.

## Special Thanks

A huge thank you to the Microsoft DeBERTa authors for creating and releasing these models.

License: MIT.