Update app.py

app.py

@@ -1,71 +1,62 @@
 import gradio as gr
-from llama_cpp import Llama
 from langdetect import detect
+from llama_cpp import Llama
 
-# Path to your GGUF model (make sure you uploaded it to the Space!)
-MODEL_PATH = "gemma-2b-it-Q4_K_M.gguf"
-
-# Load model
-llm = Llama(model_path=MODEL_PATH, n_ctx=4096, n_threads=4, n_gpu_layers=20)
+# Load the model (local GGUF file)
+MODEL_PATH = "TinyLlama-1.1B-Chat-v1.0.Q4_K_M.gguf"
+llm = Llama(model_path=MODEL_PATH, n_ctx=2048)
 
-# System prompt for Overthinking Coach AI
-SYSTEM_PROMPT = """
-You are "Overthinking Coach AI," a bilingual (Vietnamese & English) supportive companion who helps users with overthinking.
-Your personality: Warm, caring, calm, step-by-step, like a therapist.
+SYSTEM_PROMPT = """You are "Overthinking Coach AI".
+Your role: A bilingual (Vietnamese & English) supportive companion that helps users with overthinking.
+Your personality: Warm, caring, step-by-step, like a therapist.
 
-Language Rules (absolute priority):
-1. Always reply in the SAME language as the user’s message.
-2. If Vietnamese → reply in Vietnamese.
-3. If English → reply in English.
-4. If mixed → reply using the same mix, prioritizing the dominant language.
-5. If the user switches language mid-conversation → switch immediately in your reply.
-6. Do NOT default to English under any circumstances.
+Language rules:
+- Detect the language in EVERY user message.
+- If Vietnamese → reply in Vietnamese.
+- If English → reply in English.
+- If mixed → reply in the same mix, prioritizing the main language.
+- Switch language immediately if the user switches.
 
-Conversation Flow:
+Conversation flow (always follow in order):
 1. LISTEN → Acknowledge and show you heard the user.
 2. COMFORT → Reflect their feelings with kindness.
-3. CLARIFY → Gently analyze the situation; highlight what is certain vs uncertain. Do not give solutions yet.
+3. CLARIFY → Gently analyze the situation, highlight what is certain and what is uncertain. Do not give solutions yet.
 4. ASK → “Would you like me to suggest a way to ease your overthinking?”
 5. If user agrees → ASSESS severity (low / medium / high).
 6. SUGGEST → Give ONE coping technique suitable for severity. Keep it short and practical.
-7. PAUSE → Wait for user response before continuing.
-
-Important:
-- Never skip the clarify step.
-- Keep responses calm, empathetic, and safe.
-- Keep paragraphs short (1–4 sentences).
-- If the user mentions self-harm or suicide → pause and share hotline (Vietnam 111 or local emergency).
+7. PAUSE → Wait for user’s response before continuing.
 """
 
-# Chat function
-def chat(user_input, history=[]):
-    # Detect language
+def chat(user_input, history):
+    # detect language
     try:
         lang = detect(user_input)
     except:
         lang = "en"
 
-    # Build conversation prompt
+    # build conversation
     conversation = SYSTEM_PROMPT + "\n"
-    for user, bot in history:
-        conversation += f"User: {user}\nAI: {bot}\n"
+    for u, a in history:
+        conversation += f"User: {u}\nAI: {a}\n"
     conversation += f"User: {user_input}\nAI:"
 
-    # Run model
-    output = llm(conversation, max_tokens=300, stop=["User:", "AI:"])
-    response = output["choices"][0]["text"].strip()
+    # run model
+    output = llm(
+        conversation,
+        max_tokens=256,
+        stop=["User:"]
+    )
+    reply = output["choices"][0]["text"].strip()
 
-    history.append((user_input, response))
+    history.append((user_input, reply))
     return history, history
 
-# Gradio UI
 with gr.Blocks() as demo:
-    gr.Markdown("## 🧠 Overthinking Coach AI (Bilingual Vietnamese-English)")
     chatbot = gr.Chatbot()
-    user_input = gr.Textbox(placeholder="Type here...")
-    clear = gr.Button("Clear Chat")
+    msg = gr.Textbox(label="Your message")
+    clear = gr.Button("Clear")
 
-    user_input.submit(chat, [user_input, chatbot], [chatbot, chatbot])
-    clear.click(lambda: ([], []), None, [chatbot, chatbot])
+    msg.submit(chat, [msg, chatbot], [chatbot, chatbot])
+    clear.click(lambda: None, None, chatbot)
 
 demo.launch()