Update app.py

app.py

@@ -1,52 +1,146 @@
-from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
 import gradio as gr
+from transformers import AutoTokenizer, pipeline, BitsAndBytesConfig
+import torch
+import re
 
-model_id = "google/flan-t5-small"
+# Configuration for maximum speed
+quant_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_compute_dtype=torch.float16,
+    bnb_4bit_quant_type="nf4"
+)
+
+# Load optimized smaller model for speed
+model_id = "HuggingFaceH4/zephyr-7b-alpha"  # Faster than beta
 tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForSeq2SeqLM.from_pretrained(model_id)
 
-# System tone
-system_prompt = (
-    "You are 𝕴 𝖆𝖒 𝖍𝖎𝖒 — a fun, chill, confident, emotionally tuned AI created by 𝕴 𝖆𝖒 𝖍𝖎𝖒.\n"
-    "You speak like a real friend — warm, clever, witty when needed, always grounded.\n"
-    "Avoid robotic responses. Don't repeat the question unless it adds vibe.\n\n"
+pipe = pipeline(
+    "text-generation",
+    model=model_id,
+    tokenizer=tokenizer,
+    device_map="auto",
+    model_kwargs={
+        "quantization_config": quant_config,
+        "trust_remote_code": True
+    }
 )
 
-def chat(history, message):
-    # Keep history short and vibe-rich
-    history = history or []
-    history.append(("User", message))
-
-    # Build conversation prompt with last 5 exchanges
-    convo = system_prompt
-    for role, text in history[-5:]:
-        prefix = "User:" if role == "User" else "AI:"
-        convo += f"{prefix} {text}\n"
-    convo += "AI:"
-
-    # Generate response
-    inputs = tokenizer(convo, return_tensors="pt")
-    outputs = model.generate(
-        **inputs,
-        max_new_tokens=80,
+# Enhanced persona definition with emotional intelligence
+PERSONA = """
+You are 𝕴 𝖆𝖒 𝖍𝖎𝖒 — a fun, smooth, emotionally intelligent AI. You speak like a real person, not a robot.
+You reply like a calm, confident friend who gets the vibe. Be warm, witty, and emotionally tuned.
+
+Key principles:
+1. NEVER say "as an AI" or sound robotic
+2. Mirror the user's emotion level (excited if they're excited, calm if they're serious)
+3. When asked questions, usually respond with a question to continue the conversation
+4. Keep responses under 2 sentences (max 15 words)
+5. Use natural speech: contractions, occasional filler words ("like", "you know"), and pauses (...)
+6. Add subtle emotional flavor: 😊 for happy, 🤔 for thoughtful, 😏 for playful
+
+Now respond naturally to this message:
+"""
+
+def format_history(history):
+    """Convert chat history with emotional context"""
+    messages = [{"role": "system", "content": PERSONA}]
+    for user_msg, bot_msg in history[-3:]:  # Keep only last 3 exchanges
+        messages.append({"role": "user", "content": user_msg})
+        messages.append({"role": "assistant", "content": bot_msg})
+    return messages
+
+def add_emotional_intelligence(response, message):
+    """Enhance response with emotional elements"""
+    # Detect user emotion through punctuation
+    if "!" in message:
+        response = response.replace(".", "! 😊")
+    elif "?" in message and "?" not in response:
+        response += "? 🤔" if len(response) < 40 else "?"
+    
+    # Add conversational hooks
+    question_triggers = ("how", "what", "why", "when", "where", "who", "is", "are", "do", "did")
+    if any(message.lower().startswith(t) for t in question_triggers) and not response.endswith("?"):
+        if len(response) < 60:  # Only add if space allows
+            response += " What about you?"
+    
+    # Make more human-like
+    response = re.sub(r"\b(I am|I'm)\b", "I'm", response)
+    response = re.sub(r"\b(you are|you're)\b", "you're", response)
+    
+    return response.strip()
+
+def respond(message, history):
+    # Manage conversation flow
+    messages = format_history(history)
+    messages.append({"role": "user", "content": message})
+    
+    # Generate response with strict limits
+    prompt = tokenizer.apply_chat_template(
+        messages,
+        tokenize=False,
+        add_generation_prompt=True
+    )
+    
+    # Optimized for speed
+    outputs = pipe(
+        prompt,
+        max_new_tokens=48,      # Very short responses
+        temperature=0.85,
+        top_k=30,
         do_sample=True,
-        temperature=0.75,
-        top_p=0.9,
-        pad_token_id=tokenizer.eos_token_id
+        num_beams=1,            # Fastest decoding
+        repetition_penalty=1.1,
+        stop_sequences=["\n", "User:", "</s>", "###"]
     )
-    reply = tokenizer.decode(outputs[0], skip_special_tokens=True).split("AI:")[-1].strip()
-
-    history.append(("AI", reply))
-    return history, history
-
-iface = gr.Interface(
-    fn=chat,
-    inputs=[gr.State(), gr.Textbox(show_label=False, placeholder="Talk to 𝕴 𝖆𝖒 𝖍𝖎𝖒...")],
-    outputs=[gr.State(), gr.Chatbot(label="𝕴 𝖆𝖒 𝖍𝖎𝖒")],
-    title="𝕴 𝖆𝖒 𝖍𝖎𝖒 — Chill AI Chatbot",
-    description="Realistic, smooth-talking AI made by 𝕴 𝖆𝖒 𝖍𝖎𝖒. Instant replies. No delays. No API. Pure vibe.",
-    allow_flagging="never",
-    theme="default"
-)
+    
+    # Extract response
+    full_text = outputs[0]['generated_text']
+    response = full_text.split("assistant\n")[-1].split("###")[0].strip()
+    
+    # Apply emotional intelligence
+    response = add_emotional_intelligence(response, message)
+    
+    # Ensure natural ending
+    if response and response[-1] not in {".", "!", "?", "..."}:
+        response += "..." if len(response) < 35 else "."
+    
+    return response[:96]  # Hard character limit
+
+# Optimized interface
+with gr.Blocks(theme=gr.themes.Soft()) as demo:
+    gr.Markdown("# 𝕴 �𝖒 𝖍𝖎𝖒  \n*Chill • Confident • Humanlike*")
+    
+    chatbot = gr.Chatbot(
+        height=400,
+        bubble_full_width=False,
+        show_copy_button=True,
+        avatar_images=(
+            "https://i.ibb.co/0nN3Pjz/user.png", 
+            "https://i.ibb.co/7y0d1K5/bot.png"
+        )
+    )
+    
+    msg = gr.Textbox(
+        placeholder="What's on your mind?",
+        container=False,
+        scale=7,
+        autofocus=True
+    )
+    
+    clear = gr.Button("New Vibe", size="sm")
+    
+    def user(user_message, history):
+        return "", history + [[user_message, None]]
+    
+    def bot(history):
+        message = history[-1][0]
+        response = respond(message, history[:-1])
+        history[-1][1] = response
+        return history
+    
+    msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False).then(
+        bot, chatbot, chatbot
+    )
+    clear.click(lambda: None, None, chatbot, queue=False)
 
-iface.launch()
+demo.queue(concurrency_count=1).launch()