README.md

---
pipeline_tag: feature-extraction
license: apache-2.0
base_model: google/electra-small-discriminator
library_name: kerasformers
tags:
- keras
- kerasformers
- electra
- discriminator
- text-encoder
- feature-extraction
- arxiv:2003.10555
- pytorch
- jax
- tf
---

## ***See [our collection](https://huggingface.co/collections/kerasformers/electra-6a8540d1f5831e07dc89d8d1) for all versions of ELECTRA.***

# Run ELECTRA 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-ELECTRA-blue)](https://imvision12.github.io/KerasFormers/electra/) [![Collection](https://img.shields.io/badge/HF-ELECTRA%20collection-yellow)](https://huggingface.co/collections/kerasformers/electra-6a8540d1f5831e07dc89d8d1)

# kerasformers/electra_small_discriminator

Paper: [ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators (arXiv:2003.10555)](https://arxiv.org/abs/2003.10555) · [HF Papers](https://huggingface.co/papers/2003.10555)

ELECTRA is Google's BERT-style bidirectional text encoder, pre-trained as a replaced-token **discriminator** (with a smaller **generator** producing the corrupted tokens). This repo is the **encoder / downstream** checkpoint. WordPiece tokenizer; mask token `[MASK]`.

For more details on the model, please go to the upstream [model card](https://huggingface.co/google/electra-small-discriminator).

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

## ✨ Quick start (encoder / downstream)

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

from kerasformers.models.electra import ElectraModel, ElectraTokenizer

model = ElectraModel.from_weights("kerasformers/electra_small_discriminator")
tokenizer = ElectraTokenizer.from_weights("kerasformers/electra_small_discriminator")

out = model(tokenizer("The quick brown fox."))["last_hidden_state"]  # (1, L, H)
```

The same repo also serves the task heads, loaded the same way: `ElectraSequenceClassify`, `ElectraTokenClassify`, `ElectraQnA`, `ElectraMultipleChoice` (each takes the pretrained encoder and a randomly-initialized head, ready for fine-tuning).

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

| Size | Discriminator (encoder / downstream) | Generator (masked-LM) |
|---|---|---|
| small | [`kerasformers/electra_small_discriminator`](https://huggingface.co/kerasformers/electra_small_discriminator) | [`kerasformers/electra_small_generator`](https://huggingface.co/kerasformers/electra_small_generator) |
| base | [`kerasformers/electra_base_discriminator`](https://huggingface.co/kerasformers/electra_base_discriminator) | [`kerasformers/electra_base_generator`](https://huggingface.co/kerasformers/electra_base_generator) |
| large | [`kerasformers/electra_large_discriminator`](https://huggingface.co/kerasformers/electra_large_discriminator) | [`kerasformers/electra_large_generator`](https://huggingface.co/kerasformers/electra_large_generator) |

## Available classes

Load any of these from this repo with `from_weights("kerasformers/electra_small_discriminator")` (or on the fly via the `hf:` prefix). The pretrained backbone is shared; task heads not stored in this checkpoint start randomly initialized, ready for fine-tuning (or load a `hf:` fine-tune).

| Class | Task |
|---|---|
| `ElectraModel` | Encoder backbone |
| `ElectraSequenceClassify` | Sequence classification |
| `ElectraTokenClassify` | Token classification (NER / POS) |
| `ElectraQnA` | Extractive question answering |
| `ElectraMultipleChoice` | Multiple choice |

```python
from kerasformers.models.electra import ElectraSequenceClassify
model = ElectraSequenceClassify.from_weights("kerasformers/electra_small_discriminator")
```

## Tips

- Set `KERAS_BACKEND` **before** importing Keras / kerasformers.
- Prefer `ElectraTokenizer.from_weights(...)` so WordPiece tokenization matches.
- Downstream tasks (classification / QA / NER) use the **discriminator** repos; the **generator** repos are the masked-LM.
- See [ELECTRA docs](https://imvision12.github.io/KerasFormers/electra/) and [Loading Weights](https://imvision12.github.io/KerasFormers/loading_weights/).
- Community / upstream safetensors still work via the `hf:` prefix, e.g. `ElectraModel.from_weights("hf:google/electra-small-discriminator")`.

## Special Thanks

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

License: Apache 2.0.