app.py

import faiss
import gradio as gr
import numpy as np
import pandas as pd
import torch.nn.functional as F
from sentence_transformers import SentenceTransformer

DIM = 768
model = SentenceTransformer("nomic-ai/nomic-embed-text-v1.5", trust_remote_code=True)

papers_df = pd.read_csv("data/cvpr2024_papers_with_details.csv", index=None, on_bad_lines='skip')
papers_df = papers_df[~papers_df["summary"].isna() & ~papers_df["pdf_path"].isna()]

with open('data/embeddings.npy', 'rb') as f:
    embeddings = np.load(f)

index = faiss.IndexFlatL2(DIM)
index.add(embeddings)


def encode_query(query):
    query_embeddings = model.encode([query], convert_to_tensor=True)
    query_embeddings = F.layer_norm(embeddings, normalized_shape=(embeddings.shape[1],))
    query_embeddings = embeddings[:, :DIM]
    query_embeddings = F.normalize(embeddings, p=2, dim=1)
    return query_embeddings

def search_nearest_papers(query, k=5):
    query_embeddings = encode_query(query)
    D, I = index.search(query_embeddings.numpy(), k)
    return papers_df.iloc[I[0]][["title", "summary", "pdf_path"]]

demo = gr.Interface(
    search_nearest_papers,
    [
        "text",
        gr.inputs.Slider(1, 10, default=5),
    ],
    "dataframe",
)

if __name__ == "__main__":
    demo.launch()