Update app.py

app.py

@@ -36,14 +36,14 @@ TOPICS = [
     "Smart Contracts",
     "Versioning in Plutus",
     "Monad",
-    "Other"
+    "Other"  # Add more as needed
 ]
 
-# Improved prompt template for autism-friendly, focused response
+# Prompt template to guide the model
 def create_prompt(personality, level, topic):
-    return f"Explain {topic} in Plutus for a {level} programmer with {personality} traits. Use only basic words and clear examples, like comparing validation to a lock. Avoid jargon like 'blockchain,' 'ledger,' 'Haskell,' 'decentralized,' 'formal verification,' or 'immutability.' Use short sentences (8 words or less). Use exactly 3 bullet points for key ideas. Each point must be under 15 words. Include one simple analogy. Structure the response: 2-sentence introduction, 3 bullet points, 1-sentence conclusion. For Autistic traits, use literal language, avoid abstract terms, and ensure a predictable format. Do not repeat the topic or prompt. Do not simulate a conversation, ask questions, or list unrelated terms. Use a direct, instructional tone without 'I,' 'we,' or conversational phrases. End with a summary sentence on the topic’s importance. Add extra line breaks between sections for readability."
+    return f"User: Teach me about {topic} in Plutus. I am a {level} programmer with {personality} traits. Make the explanation tailored to my needs, easy to understand, and engaging.\nAssistant:"
 
-# Response function with improved parameters
+# Response function with proper streaming
 def generate_response(personality, level, topic):
     try:
         logger.info("Processing selections...")
@@ -56,16 +56,15 @@ def generate_response(personality, level, topic):
         generation_kwargs = {
             **inputs,
             "streamer": streamer,
-            "max_new_tokens": 400,  # Increased to avoid cut-offs
+            "max_new_tokens": 500,
             "do_sample": True,
-            "temperature": 0.2,     # Lowered for more focused output
-            "top_p": 0.2,          # Lowered for more focused output
-            "repetition_penalty": 1.5,  # Increased to prevent repetition
+            "temperature": 0.4,
+            "top_p": 0.5,
             "eos_token_id": tokenizer.eos_token_id,
             "pad_token_id": tokenizer.pad_token_id
         }
         
-        # Run generation in a separate thread
+        # Run generation in a separate thread to avoid blocking
         thread = Thread(target=model.generate, kwargs=generation_kwargs)
         thread.start()
         
@@ -87,17 +86,17 @@ with gr.Blocks(title="Cardano Plutus AI Assistant") as demo:
     personality = gr.Dropdown(
         choices=PERSONALITY_TYPES,
         label="Personality Type",
-        value="Autistic"
+        value="Dyslexic"  # Default
     )
     level = gr.Dropdown(
         choices=PROGRAMMING_LEVELS,
         label="Programming Level",
-        value="Beginner"
+        value="Beginner"  # Default
     )
     topic = gr.Dropdown(
         choices=TOPICS,
         label="Topic",
-        value="Introduction to Validation"
+        value="What is Plutus"  # Default
     )
     
     generate_btn = gr.Button("Generate")