Update app.py

app.py

@@ -1,18 +1,23 @@
-from fastapi import FastAPI, Request, Form
+import os
 import torch
+import gradio as gr
+from fastapi import FastAPI, Request, Form
 from transformers import AutoTokenizer, AutoModelForCausalLM
 import re
+import time
+
+# Create writable cache directory
+os.makedirs("/tmp/cache", exist_ok=True)
+os.environ["TRANSFORMERS_CACHE"] = "/tmp/cache"
+os.environ["HF_HOME"] = "/tmp/cache"
 
 app = FastAPI()
 
-# Load model
+# Lightweight CPU model
 model_id = "microsoft/DialoGPT-medium"
 tokenizer = AutoTokenizer.from_pretrained(model_id)
 model = AutoModelForCausalLM.from_pretrained(model_id)
 
-# Chat memory storage
-chat_memories = {}
-
 # Persona definition
 PERSONA = """
 [System: You are 𝕴 𝖆𝖒 𝖍𝖎𝖒 - a fun, smooth, emotionally intelligent AI. 
@@ -20,32 +25,52 @@ You speak like a real person, not a robot. Reply like a calm, confident friend w
 Keep responses under 15 words. Use natural speech. Add emotional flavor: 😊 🤔 😏]
 """
 
+# Chat memory storage
+chat_memories = {}
+
 def format_context(history):
+    """Create context using last 3 exchanges"""
     context = PERSONA + "\n"
-    for user, bot in history[-3:]:
+    
+    # Add last 3 exchanges
+    for exchange in history[-3:]:
+        user, bot = exchange
         context += f"You: {user}\n"
         context += f"𝕴 𝖆𝖒 𝖍𝖎𝖒: {bot}\n"
     return context
 
 def add_emotional_intelligence(response, message):
+    """Enhance response with emotional elements"""
+    # Add emoji based on content
     if "!" in message or any(w in response.lower() for w in ["cool", "great", "love", "awesome"]):
         response += " 😊"
     elif "?" in message or any(w in response.lower() for w in ["think", "why", "how", "consider"]):
         response += " 🤔"
     
+    # Add conversational hooks
     if "?" in message and not response.endswith("?"):
         if len(response.split()) < 10:
             response += " What do you think?"
     
+    # Make more human-like
     response = response.replace("I am", "I'm").replace("You are", "You're")
+    
+    # Free-tier: Limit to 15 words max
     words = response.split()
-    return " ".join(words[:15]) + "..." if len(words) > 15 else response
+    if len(words) > 15:
+        response = " ".join(words[:15]) + "..."
+    
+    return response
 
 def generate_response(message, session_id):
+    """Generate response with memory context"""
     history = chat_memories.get(session_id, [])
     context = format_context(history) + f"You: {message}\n𝕴 𝖆𝖒 𝖍𝖎𝖒:"
     
+    # Tokenize for CPU efficiency
     inputs = tokenizer.encode(context, return_tensors="pt")
+    
+    # Generate response
     outputs = model.generate(
         inputs,
         max_new_tokens=48,
@@ -57,21 +82,25 @@ def generate_response(message, session_id):
         pad_token_id=tokenizer.eos_token_id
     )
     
+    # Decode and extract response
     full_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
     response = full_text.split("𝕴 𝖆𝖒 𝖍𝖎𝖒:")[-1].strip()
     
+    # Clean extra dialog
     if "\nYou:" in response:
         response = response.split("\nYou:")[0]
     
+    # Apply emotional intelligence
     response = add_emotional_intelligence(response, message)
     
+    # Ensure natural ending
     if response and response[-1] not in {".", "!", "?", "..."}:
         response += "." if len(response) > 20 else "..."
     
     # Update chat history
     chat_memories[session_id] = history + [[message, response]]
     
-    return response[:80]
+    return response[:80]  # Hard character limit
 
 # API Endpoint
 @app.post("/chat")
@@ -85,8 +114,6 @@ async def chat_api(
 
 # Gradio Interface
 if __name__ == "__main__":
-    import gradio as gr
-    
     with gr.Blocks(title="𝕴 𝖆𝖒 𝖍𝖎𝖒", theme=gr.themes.Soft()) as demo:
         session_state = gr.State("default")
         gr.Markdown("# 𝕴 𝖆𝖒 𝖍𝖎𝖒 Chat API")