uncased_L-4_H-768_A-12

BERT uncased_L-4_H-768_A-12

This model is a PyTorch conversion of the original TensorFlow BERT checkpoint.

Model Details

• Model Type: BERT (Bidirectional Encoder Representations from Transformers)
• Language: English (uncased)
• Architecture:
  • Layers: 4
  • Hidden Size: 768
  • Attention Heads: 12
  • Vocabulary Size: 30522
  • Max Position Embeddings: 512

Model Configuration

{
  "hidden_size": 768,
  "hidden_act": "gelu",
  "initializer_range": 0.02,
  "vocab_size": 30522,
  "hidden_dropout_prob": 0.1,
  "num_attention_heads": 12,
  "type_vocab_size": 2,
  "max_position_embeddings": 512,
  "num_hidden_layers": 4,
  "intermediate_size": 3072,
  "attention_probs_dropout_prob": 0.1
}

Usage

from transformers import BertForPreTraining, BertTokenizer

# Load the model and tokenizer
model = BertForPreTraining.from_pretrained('bansalaman18/uncased_L-4_H-768_A-12')
tokenizer = BertTokenizer.from_pretrained('bansalaman18/uncased_L-4_H-768_A-12')

# Example usage
text = "Hello, this is a sample text for BERT."
inputs = tokenizer(text, return_tensors='pt')
outputs = model(**inputs)

Training Data

This model was originally trained on the same data as the standard BERT models:

• English Wikipedia (2500M words)
• BookCorpus (800M words)

Conversion Details

This model was converted from the original TensorFlow checkpoint to PyTorch format using a custom conversion script with the Hugging Face Transformers library.

Citation

@article{devlin2018bert,
  title={BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding},
  author={Devlin, Jacob and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
  journal={arXiv preprint arXiv:1810.04805},
  year={2018}
}