AventIQ-AI/all-MiniLM-L6-v2-book-recommendation-system

Book Recommendation System with Sentence Transformers (all-MiniLM-L6-v2)

📌 Overview

This repository hosts the quantized version of the all-MiniLM-L6-v2 model fine-tuned for book reccommendation tasks. The model has been trained on the Book_Genre dataset from Hugging Face. The model is quantized to Float16 (FP16) to optimize inference speed and efficiency while maintaining high performance.

🏗 Model Details

• Model Architecture: all-MiniLM-L6-v2
• Task: Book Recommendation System
• Dataset: Hugging Face's Book_Genre
• Quantization: Float16 (FP16) for optimized inference
• Fine-tuning Framework: Hugging Face Transformers

🚀 Usage

Installation

pip install transformers torch

Loading the Model

from sentence_transformers import SentenceTransformer, models
from huggingface_hub import hf_hub_download
from sentence_transformers.util import cos_sim
import torch
import ast

device = "cuda" if torch.cuda.is_available() else "cpu"

model_name = "AventIQ-AI/all-MiniLM-L6-v2-book-recommendation-system"
model = SentenceTransformer(model_name).to(device)

Question Answer Example

# Load the Embeddings CSV file
df_embeddings = pd.read_csv(csv_path)

# Convert embedding column (string) back to list
df_embeddings["embedding"] = df_embeddings["embedding"].apply(ast.literal_eval)

# Convert embeddings to tensor
book_embeddings = torch.tensor(df_embeddings["embedding"].tolist())

def get_book_recommendations(query, top_k=5):
    query_embedding = model.encode(query, convert_to_tensor=True)

    if query_embedding.ndim == 1:
        query_embedding = query_embedding.unsqueeze(0)

    if isinstance(book_embeddings, np.ndarray):
        book_embeddings_tensor = torch.tensor(book_embeddings, dtype=torch.float32)
    else:
        book_embeddings_tensor = book_embeddings  # If already tensor, use as is

    similarities = cos_sim(query_embedding, book_embeddings_tensor)

    similarities = similarities.squeeze(0)  # Remove extra dimension if necessary

    top_k_values, top_k_indices = torch.topk(similarities, k=top_k)

    recommended_titles = df_embeddings.iloc[top_k_indices.cpu().numpy()]['title'].tolist()
    recommended_scores = top_k_values.cpu().numpy().tolist()

    return list(zip(recommended_titles, recommended_scores))

# Example usage:
query = "A horror novel with ghosts and dark nights"
recommendations = get_book_recommendations(query, top_k=5)

print("Book Recommendations:")
for title, score in recommendations:
    print(f"{title} - Score: {score:.4f}")

⚡ Quantization Details

Post-training quantization was applied using PyTorch's built-in quantization framework. The model was quantized to Float16 (FP16) to reduce model size and improve inference efficiency while balancing accuracy.

Evaluation Metrics: NDCG

NDCG → If close to 1, the ranking matches expected relevance. Our model's NDCG score is 0.82

🔧 Fine-Tuning Details

Dataset

The Bookcorpus dataset was used for training and evaluation. The dataset consists of texts.

Training Configuration

• Number of epochs: 3
• warmup steps: 100
• Evaluation strategy: steps

📂 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 struggle for out of scope tasks.
• Quantization may lead to slight degradation in accuracy compared to full-precision models.
• Performance may vary across different writing styles and sentence structures.

🤝 Contributing

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