app.py

import pickle
import os
import torch
print(os.getcwd())

# Load embeddings
fileobj = open("/home/user/app/embmmn7.obj", "rb")
corpus_embeddings, corpus = pickle.load(fileobj)
fileobj.close()

from sentence_transformers import SentenceTransformer, util
embedder = SentenceTransformer("ramdane/jurimodel")

# Add normalize_embeddings function
def normalize_embeddings(embeddings):
    """Normalize embeddings to unit length, handling both 1D and 2D tensors"""
    if isinstance(embeddings, torch.Tensor):
        # Check if it's a 1D tensor (single embedding)
        if len(embeddings.shape) == 1:
            # Add batch dimension, normalize, then squeeze
            return torch.nn.functional.normalize(embeddings.unsqueeze(0), p=2, dim=1).squeeze(0)
        else:
            # It's already 2D, normalize as usual
            return torch.nn.functional.normalize(embeddings, p=2, dim=1)
    else:
        # Convert to tensor first
        embeddings_tensor = torch.tensor(embeddings)
        if len(embeddings_tensor.shape) == 1:
            return torch.nn.functional.normalize(embeddings_tensor.unsqueeze(0), p=2, dim=1).squeeze(0)
        else:
            return torch.nn.functional.normalize(embeddings_tensor, p=2, dim=1)

# Ensure corpus embeddings are normalized when loaded
if not isinstance(corpus_embeddings, torch.Tensor):
    corpus_embeddings = torch.tensor(corpus_embeddings)
corpus_embeddings = normalize_embeddings(corpus_embeddings)

def showr(queries, number):
    # Create query embedding
    query_embedding = embedder.encode(queries, convert_to_tensor=True)
    
    # Normalize query embedding
    query_embedding = normalize_embeddings(query_embedding)
    
    # Perform semantic search
    hits = util.semantic_search(query_embedding, corpus_embeddings, top_k=10)
    hits = hits[0]  # Get the hits for the first query
    
    if hits[number]['score'] > 0.05:
        return corpus[hits[number]['corpus_id']]
    else:
        return "لم نتمكن من ايجاد النتيجة اما لعدم وجود الاجتهاد او لعدم كتابة جملة بحث مناسبة "

import gradio as gr

def greet(search_for, number):
    return showr(search_for, int(number))

iface = gr.Interface(
    fn=greet, 
    inputs=[
        gr.Textbox(label="ادخل كلمات البحث"),
        gr.Number(label="الترتيب")
    ], 
    outputs=gr.TextArea(label="الاجتهاد")
)

iface.launch()