DistilBERT-Base-Uncased Quantized Model for Scientific Paper Classification

This repository hosts a quantized version of the DistilBERT model, fine-tuned for scientific paper classification into three categories: Biology, Mathematics, and Physics. The model has been optimized for efficient deployment while maintaining high accuracy, making it suitable for real-world applications, including academic research and automated categorization of scientific literature.

Model Details

Model Architecture: DistilBERT Base Uncased
Task: Scientific Paper Classification
Dataset: Custom dataset labeled with three categories: Biology, Mathematics, and Physics
Quantization: Float16 (FP16)
Fine-tuning Framework: Hugging Face Transformers

Usage

Installation

pip install transformers torch

Loading the Model

from transformers import DistilBertForSequenceClassification, DistilBertTokenizer
import torch

# Load quantized model
quantized_model_path = "/kaggle/working/distilbert_finetuned_fp16"
quantized_model = DistilBertForSequenceClassification.from_pretrained(quantized_model_path)
quantized_model.eval()  # Set to evaluation mode
quantized_model.half()  # Convert model to FP16

# Load tokenizer
tokenizer = DistilBertTokenizer.from_pretrained("distilbert-base-uncased")

# Define a test input
test_paper = "The quantum mechanics of atomic structures are governed by Schrödinger's equation."

# Tokenize input
inputs = tokenizer(test_paper, return_tensors="pt", padding=True, truncation=True, max_length=512)

# Ensure input tensors are in correct dtype
inputs["input_ids"] = inputs["input_ids"].long()  # Convert to long type
inputs["attention_mask"] = inputs["attention_mask"].long()  # Convert to long type

# Make prediction
with torch.no_grad():
    outputs = quantized_model(**inputs)

# Get predicted class
predicted_class = torch.argmax(outputs.logits, dim=1).item()

# Class labels
label_mapping = {0: "Biology", 1: "Mathematics", 2: "Physics"}

predicted_label = label_mapping[predicted_class]
print(f"Predicted Label: {predicted_label}")

Performance Metrics

Accuracy: 0.95 (after fine-tuning)
F1-Score: 0.91 (weighted)

Fine-Tuning Details

Dataset

The dataset consists of scientific papers categorized into three domains:

Biology
Mathematics
Physics

The dataset was preprocessed and tokenized using the DistilBERT tokenizer.

Training

Number of epochs: 3
Batch size: 8
Learning rate: 2e-5
Optimizer: AdamW
Evaluation strategy: epoch

Quantization

Post-training quantization was applied using PyTorch’s built-in quantization framework to reduce the model size and improve inference efficiency.

Repository Structure

.
├── model/               # Contains the quantized model files
├── tokenizer_config/    # Tokenizer configuration and vocabulary files
├── model.safensors/     # Fine-Tuned Model
├── README.md            # Model documentation

Limitations

The model is trained on a limited dataset and may not generalize well to niche scientific subdomains.
Quantization may result in slight accuracy degradation compared to full-precision models.

Contributing

Contributions are welcome! Feel free to open an issue or submit a pull request if you have suggestions or improvements.