bert-finetuned-ner-accelerate

This repository contains a BERT model fine-tuned for Named Entity Recognition (NER) using the 🤗 Accelerate library for efficient training and evaluation.

---

Model Overview

• Base model: BERT (pretrained)
• Task: Named Entity Recognition (Token Classification)
• Fine-tuning framework: Hugging Face Transformers + Accelerate
• Optimizer: AdamW with learning rate 2e-5
• Learning rate scheduler: Linear scheduler with no warmup steps
• Training epochs: 3
• Batch training with multi-GPU/TPU support via Accelerate

---

Training Details

• Optimizer used: AdamW from PyTorch with lr=2e-5.
• Learning rate scheduler: Linear decay over total training steps.
• Total training steps: num_train_epochs * len(train_dataloader).
• Training and evaluation are done inside a loop over epochs with progress bar tracking.

---

Evaluation Metrics

Epoch	Precision	Recall	F1 Score	Accuracy
0	0.9423	0.9239	0.9330	0.9848
1	0.9487	0.9258	0.9371	0.9862
2	0.9487	0.9258	0.9371	0.9862

The metrics are calculated on the evaluation dataset after each epoch.

---

Usage

Load the model and tokenizer using the Hugging Face transformers library:

from transformers import AutoTokenizer, AutoModelForTokenClassification

model_name = "Ak128umar/bert-finetuned-ner-accelerate"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)