KerasNLP Nightly

0.30.0.dev202605080526 verified Sat May 09 auth: no python

Nightly build of KerasNLP, providing pretrained NLP models and tokenizers for Keras 3. Current version 0.30.0.dev202605080526. Released daily with latest changes from keras-team/keras/keras_hub.

pip install keras-nlp-nightly

Common errors

error ModuleNotFoundError: No module named 'keras_nlp' ↓

cause Missing installation or wrong package name.

fix

pip install keras-nlp-nightly

error AttributeError: module 'keras_nlp' has no attribute 'BertClassifier' ↓

cause Importing from top-level package instead of subpackage.

fix

from keras_nlp.models import BertClassifier

error ValueError: Unknown preset: 'bert_base_uncased' ↓

cause Preset may have been removed or renamed in nightly.

fix

List presets: keras_nlp.presets.list() or check docs.

Warnings

breaking KerasNLP has been merged into Keras 3 (keras-core) since Keras 3.0. Nightly builds may break without notice. ↓

fix Use from keras_nlp import ... as shown in quickstart. For production, use stable KerasNLP (pip install keras-nlp).

deprecated Some presets and symbols may be removed or renamed in nightly builds before stable release. ↓

fix Check changelog at https://github.com/keras-team/keras/keras_hub

gotcha Mixing keras-nlp-nightly with stable keras may cause version conflicts due to incompatible keras dependencies. ↓

fix Use consistent versions: either both stable or both nightly. Prefer using within a virtualenv.

Imports

BertClassifier

wrong

from keras_nlp import BertClassifier

correct

from keras_nlp.models import BertClassifier

Wrong due to subpackage structure

Tokenizer

wrong

import Tokenizer from keras_nlp

correct

from keras_nlp.tokenizers import Tokenizer

Tokenizer is in subpackage

Quickstart

Quickstart: import keras_nlp, set backend, load a model. Use from_preset with a valid preset name.

import os
os.environ['KERAS_BACKEND'] = 'jax'
import keras_nlp
from keras_nlp.models import BertClassifier

# Load a pretrained classifier
classifier = BertClassifier.from_preset("bert_tiny_en_uncased")
print(classifier.summary())