BERT-Small

https://arxiv.org/abs/1810.04805

Lucid port of transformers/google/bert_uncased_L-4_H-512_A-8, converted to Lucid-native safetensors.

Available weights

Tag	Params	GFLOPs	Size	Source
WIKIPEDIA_BOOKSCORPUS (default)	28.8M	—	109.73 MB	transformers

Usage

import lucid
import lucid.models as models
from lucid.models.weights import BertSmallWeights

# default tag
model = models.bert_small(pretrained=True)

# explicit tag (enum or string)
model = models.bert_small(weights=BertSmallWeights.WIKIPEDIA_BOOKSCORPUS)
model = models.bert_small(pretrained="WIKIPEDIA_BOOKSCORPUS")

# feed token ids (tokenize with the matching lucid.utils.tokenizer)
input_ids = lucid.tensor([[101, 7592, 2088, 102]], dtype=lucid.int64)
out = model(input_ids)
hidden = out.last_hidden_state  # (B, T, hidden_size)

Conversion

Converted from transformers/google/bert_uncased_L-4_H-512_A-8 via python -m tools.convert_weights bert_small --tag WIKIPEDIA_BOOKSCORPUS. Key mapping + numerical parity verified against the source.

License

apache-2.0 — inherited from the original weights.

Citation

Devlin et al., "BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding", NAACL 2019. Miniatures: Turc et al., "Well-Read Students Learn Better", 2019.