Update utils.py

utils.py

@@ -3,19 +3,20 @@ import pandas as pd
 import streamlit as st
 from huggingface_hub import InferenceClient
 
-# Initialize Hugging Face Inference client using the secret
-hf_token = st.secrets.get("HF_TOKEN")
-if not hf_token:
-    st.error("HF_TOKEN not found in secrets. Please add it.")
-    st.stop()
-
-client = InferenceClient(token=hf_token)
+def get_hf_client():
+    hf_token = st.secrets.get("HF_TOKEN")
+    if not hf_token:
+        st.error("HF_TOKEN not found in Streamlit secrets!")
+        st.stop()
+    return InferenceClient(token=hf_token)
 
 def query_agent_from_csv(file_bytes, user_query, model_repo="mistralai/Mistral-7B-Instruct-v0.3"):
     """
     Reads a CSV and queries the Hugging Face Mistral model.
     Returns the model's answer as string.
     """
+    client = get_hf_client()  # Initialize client inside function
+
     try:
         # --- Step 1: Load CSV ---
         try:
@@ -24,18 +25,17 @@ def query_agent_from_csv(file_bytes, user_query, model_repo="mistralai/Mistral-7
             file_bytes.seek(0)
             df = pd.read_csv(file_bytes, encoding="latin1")
         
-        # Limit columns to avoid huge inputs
         MAX_COLS = 50
         if df.shape[1] > MAX_COLS:
             df = df.iloc[:, :MAX_COLS]
         
-        # --- Step 2: Summarize dataset for model context ---
+        # --- Step 2: Summarize dataset ---
         summary = f"The dataset has {df.shape[0]} rows and {df.shape[1]} columns.\n"
         summary += "Columns: " + ", ".join(df.columns[:10])
         if df.shape[1] > 10:
             summary += ", ..."
         
-        # --- Step 3: Build messages for chat API ---
+        # --- Step 3: Build messages ---
         messages = [
             {"role": "system", "content": (
                 "You are a professional data analyst. "
@@ -45,7 +45,7 @@ def query_agent_from_csv(file_bytes, user_query, model_repo="mistralai/Mistral-7
             {"role": "user", "content": f"Question: {user_query}"}
         ]
         
-        # --- Step 4: Query the model ---
+        # --- Step 4: Query model ---
         response = client.chat_completion(
             model=model_repo,
             messages=messages,