README.md

metadata

library_name: lucid
license: apache-2.0
tags:
  - question-answering
  - bert
  - lucid
datasets:
  - squad
pipeline_tag: question-answering
model-index:
  - name: bert-large-qa
    results:
      - task:
          type: question-answering
        dataset:
          name: squad
          type: squad
        metrics:
          - type: exact_match
            value: 86.9
          - type: f1
            value: 93.2

BERT-Large WWM (SQuAD v1.1)

https://arxiv.org/abs/1810.04805

Lucid port of transformers/google-bert/bert-large-uncased-whole-word-masking-finetuned-squad, converted to Lucid-native safetensors.

Available weights

Tag	exact_match	f1	Params	GFLOPs	Size	Source
SQUAD_V1 (default)	86.9	93.2	335.1M	—	1278.52 MB	transformers

Usage

import lucid
import lucid.models as models
from lucid.models.weights import BERTLargeQAWeights

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

# explicit tag (enum or string)
model = models.bert_large_qa(weights=BERTLargeQAWeights.SQUAD_V1)
model = models.bert_large_qa(pretrained="SQUAD_V1")

# 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)
start, end = out.start_logits, out.end_logits  # (B, T) each

Conversion

Converted from transformers/google-bert/bert-large-uncased-whole-word-masking-finetuned-squad via python -m tools.convert_weights bert_large_qa --tag SQUAD_V1. 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.