Update app.py

app.py

@@ -1,56 +1,66 @@
 import os
 import pandas as pd
 import matplotlib.pyplot as plt
+from transformers import AutoTokenizer, AutoModelForCausalLM
 import gradio as gr
 import torch
-from transformers import AutoTokenizer, AutoModelForCausalLM
 
-# Use TinyLlama (Open-source & lightweight)
-model_id = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
-tokenizer = AutoTokenizer.from_pretrained(model_id)
+
+# --- Load HF token from environment ---
+token = os.getenv("HF_TOKEN")
+if token is None:
+    raise ValueError("HF_TOKEN environment variable not set")
+
+# --- Use a better and faster model ---
+model_id = "HuggingFaceH4/zephyr-7b-beta"
+
+tokenizer = AutoTokenizer.from_pretrained(model_id, token=token)
 model = AutoModelForCausalLM.from_pretrained(
     model_id,
+    token=token,
     device_map="auto",
     torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32
 )
+
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 model.to(device)
 
+# --- Load GA4 CSV data ---
 def load_ga4_data():
     return pd.read_csv("synthetic_ga4_data.csv")
-
-def load_channel_data():
-    df = load_ga4_data()
-    agg = df.groupby("Channel").agg({
-        "Sessions": "sum",
-        "Transactions": "sum",
-        "Revenue": "sum"
-    })
-    agg["CPA (TRY)"] = (agg["Revenue"] / agg["Transactions"]).fillna(0)
-    return agg.reset_index()
-
-def generate_insight(prompt):
-    try:
         df = load_ga4_data()
         prompt = prompt.lower().strip()
 
+
         if "most users" in prompt:
             top_city = df.groupby("City")["Users"].sum().idxmax()
             users = df.groupby("City")["Users"].sum().max()
-            base_answer = f"{top_city} has the most users with {int(users)} total."
-        elif "best conversion" in prompt:
-            df["ConversionRate"] = df["Transactions"] / df["Sessions"]
-            conv = df.groupby("City")["ConversionRate"].mean()
-            top_city = conv.idxmax()
-            base_answer = f"{top_city} has the best average conversion rate of {conv.max():.2%}."
         else:
-            return "📌 I can currently only analyze user counts and conversion rates."
+            base_answer = "Sorry, I can only currently analyze questions about users and conversion rates."
 
-        input_text = f"Act like a digital marketing analyst. Rephrase for a report: {base_answer}"
+        input_text = f"Rephrase this like a digital marketing analyst for a business report: {base_answer}"
         inputs = tokenizer(input_text, return_tensors="pt").to(device)
-        outputs = model.generate(**inputs, max_new_tokens=100)
-        response = tokenizer.decode(outputs[0], skip_special_tokens=True)
 
-        return response
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=80,
+            do_sample=False,
+            temperature=0.3,
+            repetition_penalty=1.1
+        )
+
+        response = tokenizer.decode(outputs[0], skip_special_tokens=True).strip()
+        return response, "✅ Insight generated successfully."
+
     except Exception as e:
-        return f"⚠️ Error: {s
+        return f"⚠️ Error: {str(e)}", "❌ Insight generation failed."
+
+# --- City Performance Plot ---
+def plot_city_performance():
+        prompt = gr.Textbox(label="Your Analysis Question", placeholder="Which city has the best conversion rate?")
+        generate = gr.Button("Generate Insight")
+        output = gr.Textbox(label="AI Response", interactive=False)
+        status = gr.Textbox(label="Status", interactive=False, visible=True)
+        generate.click(fn=generate_insight, inputs=prompt, outputs=[output, status])
+
+    gr.Markdown("---\n**Available GA4 Metrics:** Users, Sessions, Transactions, Revenue, Avg Session Duration, etc.")