Update app.py

app.py

@@ -1,33 +1,35 @@
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
 import gradio as gr
 
-model_name = "facebook/mbart-large-50-many-to-many-mmt"
+# Load T0pp (instruction-tuned)
+model_name = "bigscience/T0pp"
 tokenizer = AutoTokenizer.from_pretrained(model_name)
 model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
 pipe = pipeline("text2text-generation", model=model, tokenizer=tokenizer)
 
-# Role-based prompt
+# Debater personas
 personas = {
-    "🟢 Optimist": "You are an optimist. Defend the topic with positivity. Give 2 reasons.",
-    "🔴 Pessimist": "You are a pessimist. Criticize the topic. Mention 2 problems.",
-    "🟡 Neutral": "You are a neutral thinker. Present pros and cons fairly."
+    "🟢 Optimist": "Give a hopeful, positive opinion with 2 reasons.",
+    "🔴 Pessimist": "Criticize the topic and point out 2 problems.",
+    "🟡 Neutral": "Give a fair and balanced view with pros and cons."
 }
 
+# Debate function
 def generate_debate(topic):
     responses = {}
     for label, instruction in personas.items():
-        prompt = f"{instruction}\nDebate Topic: {topic}"
-        result = pipe(prompt, max_new_tokens=120, do_sample=True, temperature=0.7)[0]["generated_text"]
+        prompt = f"You are a debater. {instruction}\nDebate Topic: {topic}"
+        result = pipe(prompt, max_new_tokens=150, temperature=0.7)[0]['generated_text']
         responses[label] = result.strip()
-    return "\n\n".join([f"**{k}**: {v}" for k, v in responses.items()])
+    return "\n\n".join([f"**{k}**:\n{v}" for k, v in responses.items()])
 
 # Gradio UI
 demo = gr.Interface(
     fn=generate_debate,
-    inputs=gr.Textbox(label="Enter a Debate Topic"),
+    inputs=gr.Textbox(label="Debate Topic"),
     outputs=gr.Markdown(),
-    title="🤖 Debate Club: Multi-Agent Edition",
-    description="A mini debate simulation using different perspectives powered by MBART. Runs on free-tier CPU."
+    title="🎤 Multi-Agent Debate Simulator (T0pp)",
+    description="Simulates a debate with multiple perspectives using the instruction-tuned BigScience T0pp model on Hugging Face 🤗."
 )
 
 demo.launch()