bert_large_uncased

See our collection for all versions of BERT.

Run BERT with Keras 3: JAX, PyTorch, or TensorFlow

zeromodels/bert_large_uncased

Paper: BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding (arXiv:1810.04805) · HF Papers

BERT is Google's bidirectional transformer text encoder, pretrained with masked LM and next-sentence prediction. WordPiece tokenizer; mask token [MASK]. Uncased variants lower-case the input; cased variants preserve case.

For more details on the model, please go to the upstream model card.

Pure-Keras 3 conversion of google-bert/bert-large-uncased for zeromodels. One implementation runs unmodified on TensorFlow / Torch / JAX.

This is a fill-mask / encoder checkpoint (BertMaskedLM, large uncased). Task heads (sequence/token classify, QA, NSP, …) load via hf: fine-tunes.

✨ Quick start (fill-mask)

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

from zeromodels.models.bert import BertMaskedLM, BertTokenizer

mlm = BertMaskedLM.from_weights("zeromodels/bert_large_uncased")
tokenizer = BertTokenizer.from_weights("zeromodels/bert_large_uncased")

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.ids_to_tokens[int(logits[0, mask].argmax())])

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

Available classes

Load any of these from this repo with from_weights("zeromodels/bert_large_uncased") (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).

from zeromodels.models.bert import BertSequenceClassify
model = BertSequenceClassify.from_weights("zeromodels/bert_large_uncased")

Tips

• Set KERAS_BACKEND before importing Keras / zeromodels.
• Prefer BertTokenizer.from_weights(...) so WordPiece casing matches.
• Use [MASK] (not <mask>).
• See BERT docs and Loading Weights.
• Community / upstream safetensors still work via the hf: prefix, e.g. BertMaskedLM.from_weights("hf:google-bert/bert-large-uncased").

Special Thanks

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

License: Apache 2.0.