Named Entity Recognition (NER) of Information Extraction

📌 Overview

🏗 Model Details

• Model Architecture: BERT Base Cased
• Task: Named Entity Recognition (NER) of Information Extraction
• Dataset: Hugging Face's CoNLL-2003
• Quantization: BrainFloat16
• Fine-tuning Framework: Hugging Face Transformers

🚀 Usage

Installation

Loading the Model

Named Entity Recognition Inference

🔹 Labeling Scheme (BIO Format)

• B-XYZ (Beginning): Indicates the beginning of an entity of type XYZ (e.g., B-PER for the beginning of a person’s name).
• I-XYZ (Inside): Represents subsequent tokens inside an entity (e.g., I-PER for the second part of a person’s name).
• O (Outside): Denotes tokens that are not part of any named entity.

def predict_entities(text, model):

    tokens = tokenizer(text, return_tensors="pt", truncation=True)
    tokens = {key: val.to(device) for key, val in tokens.items()}  # Move to CUDA

    with torch.no_grad():
        outputs = model(**tokens)
    
    logits = outputs.logits  # Extract logits
    predictions = torch.argmax(logits, dim=2)  # Get highest probability labels

    tokens_list = tokenizer.convert_ids_to_tokens(tokens["input_ids"][0])
    predicted_labels = [label_list[pred] for pred in predictions[0].cpu().numpy()]

    final_tokens = []
    final_labels = []
    for token, label in zip(tokens_list, predicted_labels):
        if token.startswith("##"):  
            final_tokens[-1] += token[2:]  # Merge subword
        else:
            final_tokens.append(token)
            final_labels.append(label)

    for token, label in zip(final_tokens, final_labels):
        if token not in ["[CLS]", "[SEP]"]:
            print(f"{token}: {label}")

# 🔍 Test Example
sample_text = "Elon Musk is the CEO of Tesla, which is based in California."
predict_entities(sample_text, model)

📊 Evaluation Results for Quantized Model

🔹 Overall Performance

• Accuracy: 97.10% ✅
• Precision: 89.52%
• Recall: 90.67%
• F1 Score: 90.09%

🔹 Performance by Entity Type

⏳ Inference Speed Metrics

• Total Evaluation Time: 15.89 sec
• Samples Processed per Second: 217.26
• Steps per Second: 27.18
• Epochs Completed: 3

Fine-Tuning Details

Dataset

📊 Training Details

• Number of epochs: 3
• Batch size: 8
• Evaluation strategy: epoch
• Learning Rate: 2e-5

⚡ Quantization

📂 Repository Structure

.
├── model/               # Contains the quantized model files
├── tokenizer_config/    # Tokenizer configuration and vocabulary files
├── model.safetensors/   # Quantized Model
├── README.md            # Model documentation

⚠️ Limitations

• The model may not generalize well to domains outside the fine-tuning dataset.
• Quantization may result in minor accuracy degradation compared to full-precision models.

🤝 Contributing