README.md

# Custom BERT NER Model

This repository contains a BERT-based Named Entity Recognition (NER) model fine-tuned on the CoNLL-2003 dataset. The model is trained to identify common named entity types such as persons, organizations, locations, and miscellaneous entities.

---

## Model Details

- **Model architecture:** BERT (bert-base-cased)
- **Task:** Token classification / Named Entity Recognition (NER)
- **Training data:** CoNLL-2003 dataset (~14,000 training samples)
- **Number of epochs:** 5
- **Framework:** Hugging Face Transformers + Datasets
- **Device:** CUDA-enabled GPU for training and inference
- **WandB:** Disabled during training

---

## Usage

You can use this model for token classification to identify named entities in your text.

### Installation

```python
pip install transformers datasets torch
```

## Load the model and tokenizer

```pyhton

from transformers import BertTokenizerFast, BertForTokenClassification
import torch

model_name_or_path = "AventIQ-AI/Custom-BERT-NER-Model"

tokenizer = BertTokenizerFast.from_pretrained(model_name_or_path)
model = BertForTokenClassification.from_pretrained(model_name_or_path)

model.to("cuda")  # or "cpu"
model.eval()
```

## Example inference

```python

text = "Hi, I am Deepak and I am living in Delhi."

tokens = tokenizer(text, return_tensors="pt").to(model.device)
outputs = model(**tokens)
predictions = torch.argmax(outputs.logits, dim=2)

labels = [model.config.id2label[p.item()] for p in predictions[0]]
for token, label in zip(tokenizer.tokenize(text), labels):
print(f"{token}: {label}")
```

## Training Details

 - Dataset: CoNLL-2003, loaded via the Hugging Face datasets library

 - Optimizer: AdamW

 - Learning Rate: 5e-5

 - Batch Size: 16

 - Max Sequence Length: 128

 - Epochs: 5

 - Evaluation: Performed on validation split (if applicable)

 - Quantization: Applied post-training for model size reduction (optional)

## Limitations

 - The model may not generalize well to unseen entity types or domains outside CoNLL-2003.

 - It can occasionally mislabel entities, especially for rare or new names.

 - A CUDA-enabled GPU is required for efficient training and inference.