# import gradio as gr # import pandas as pd # from sentence_transformers import SentenceTransformer, util # # Load files # df = pd.read_excel("IslamWeb_output.xlsx") # df2 = pd.read_excel("JordanFatwas_all.xlsx") # # Validate # for d, name in [(df, "IslamWeb"), (df2, "JordanFatwas")]: # if not {"question", "link"}.issubset(d.columns): # raise ValueError(f"❌ Missing required columns in {name}") # # Load model + encode # model = SentenceTransformer("paraphrase-multilingual-MiniLM-L12-v2") # embeddings = model.encode(df["question"].fillna('').tolist(), convert_to_tensor=True) # embeddings2 = model.encode(df2["question"].fillna('').tolist(), convert_to_tensor=True) # # Define function # def search_fatwa(query): # query_embedding = model.encode(query, convert_to_tensor=True) # scores = util.pytorch_cos_sim(query_embedding, embeddings)[0] # top_idx = int(scores.argmax()) # scores2 = util.pytorch_cos_sim(query_embedding, embeddings2)[0] # top_idx2 = int(scores2.argmax()) # return { # "question1": df.iloc[top_idx]["question"], # "link1": df.iloc[top_idx]["link"], # "question2": df2.iloc[top_idx2]["question"], # "link2": df2.iloc[top_idx2]["link"], # } # # Interface # iface = gr.Interface( # fn=search_fatwa, # inputs="text", # outputs="json", # allow_flagging="never", # title="Fatwa Search (Dual Source)", # description="Get the most relevant fatwas from both datasets" # ) # iface.launch() # import torch # import pandas as pd # from sentence_transformers import SentenceTransformer, util # import gradio as gr # model = SentenceTransformer("paraphrase-multilingual-MiniLM-L12-v2") # df = pd.read_csv("cleaned1.csv") # df2 = pd.read_csv("cleaned2.csv") # embeddings = torch.load("embeddings1.pt") # embeddings2 = torch.load("embeddings2.pt") # # def search_fatwa(data): # # query = data[0] if data else "" # # query_embedding = model.encode(query, convert_to_tensor=True) # # top_idx = int(util.pytorch_cos_sim(query_embedding, embeddings)[0].argmax()) # # top_idx2 = int(util.pytorch_cos_sim(query_embedding, embeddings2)[0].argmax()) # # return { # # "question1": df.iloc[top_idx]["question"], # # "link1": df.iloc[top_idx]["link"], # # "question2": df2.iloc[top_idx2]["question"], # # "link2": df2.iloc[top_idx2]["link"] # # } # def search_fatwa(data): # query = data[0] if isinstance(data, list) else data # if not query: # return {"question1": "", "link1": "", "question2": "", "link2": ""} # query_embedding = model.encode(query, convert_to_tensor=True) # top_idx = int(util.pytorch_cos_sim(query_embedding, embeddings)[0].argmax()) # top_idx2 = int(util.pytorch_cos_sim(query_embedding, embeddings2)[0].argmax()) # # return { # # "question1": df.iloc[top_idx]["question"], # # "link1": df.iloc[top_idx]["link"], # # "question2": df2.iloc[top_idx2]["question"], # # "link2": df2.iloc[top_idx2]["link"] # # } # result = f"""Question 1: {df.iloc[top_idx]["question"]} # Link 1: {df.iloc[top_idx]["link"]} # Question 2: {df2.iloc[top_idx2]["question"]} # Link 2: {df2.iloc[top_idx2]["link"]}""" # return result # iface = gr.Interface( # fn=search_fatwa, # inputs=[gr.Textbox(label="text", lines=3)], # outputs="text" # Changed from "json" to "text" # ) # # iface = gr.Interface(fn=search_fatwa, inputs=[gr.Textbox(label="text", lines=3)], outputs="json") # # iface = gr.Interface( # # fn=predict, # # inputs=[gr.Textbox(label="text", lines=3)], # # outputs='text', # # title=title, # # ) # iface.launch() import torch import pandas as pd from sentence_transformers import SentenceTransformer, util import gradio as gr model = SentenceTransformer("distilbert-base-multilingual-cased") # model = SentenceTransformer("paraphrase-multilingual-MiniLM-L12-v2") df = pd.read_csv("cleaned1.csv") df2 = pd.read_csv("cleaned2.csv") df3 = pd.read_csv("cleaned3.csv") embeddings = torch.load("embeddings1_1.pt") embeddings2 = torch.load("embeddings2_1.pt") embeddings3 = torch.load("embeddings3_1.pt") # embeddings = torch.load("embeddings1.pt") # embeddings2 = torch.load("embeddings2.pt") # embeddings3 = torch.load("embeddings3.pt") # Pre-extract DataFrame columns to avoid repeated iloc calls df_questions = df["question"].values df_links = df["link"].values df2_questions = df2["question"].values df2_links = df2["link"].values df3_questions = df3["question"].values df3_links = df3["url"].values def predict(text): if not text or text.strip() == "": return "No query provided" query_embedding = model.encode(text, convert_to_tensor=True) # Compute similarity scores sim_scores1 = util.pytorch_cos_sim(query_embedding, embeddings)[0] sim_scores2 = util.pytorch_cos_sim(query_embedding, embeddings2)[0] sim_scores3 = util.pytorch_cos_sim(query_embedding, embeddings3)[0] # Get top 3 values and indices in one call top3_scores1, top3_idx1 = sim_scores1.topk(3) top3_scores2, top3_idx2 = sim_scores2.topk(3) top3_scores3, top3_idx3 = sim_scores3.topk(3) # Convert to CPU once top3_idx1_cpu = top3_idx1.cpu().numpy() top3_idx2_cpu = top3_idx2.cpu().numpy() top3_idx3_cpu = top3_idx3.cpu().numpy() top3_scores1_cpu = top3_scores1.cpu().numpy() top3_scores2_cpu = top3_scores2.cpu().numpy() top3_scores3_cpu = top3_scores3.cpu().numpy() # Prepare results using pre-extracted arrays results = { "top2": [ { "question": df2_questions[idx], "link": df2_links[idx], "score": float(score) } for idx, score in zip(top3_idx2_cpu, top3_scores2_cpu) ], "top3": [ { "question": df3_questions[idx], "link": df3_links[idx], "score": float(score) } for idx, score in zip(top3_idx3_cpu, top3_scores3_cpu) ], "top1": [ { "question": df_questions[idx], "link": df_links[idx], "score": float(score) } for idx, score in zip(top3_idx1_cpu, top3_scores1_cpu) ], } return results # Match the EXACT structure of your working translation app title = "Search CSV" iface = gr.Interface( fn=predict, # Changed from search_fatwa to predict inputs=[gr.Textbox(label="text", lines=3)], outputs='json', title=title, ) iface.launch()