Update app.py

app.py

@@ -1,6 +1,7 @@
 import os
 import re
 import requests
+import torch
 import streamlit as st
 from langchain_huggingface import HuggingFaceEndpoint
 from langchain_core.prompts import PromptTemplate
@@ -8,6 +9,10 @@ from langchain_core.output_parsers import StrOutputParser
 from transformers import pipeline
 from langdetect import detect  # Ensure this package is installed
 
+# ✅ Check for GPU or Default to CPU
+device = "cuda" if torch.cuda.is_available() else "cpu"
+print(f"✅ Using device: {device}")  # Debugging info
+
 # ✅ Environment Variables
 HF_TOKEN = os.getenv("HF_TOKEN")
 if HF_TOKEN is None:
@@ -20,7 +25,7 @@ if NASA_API_KEY is None:
 # ✅ Set Up Streamlit
 st.set_page_config(page_title="HAL - NASA ChatBot", page_icon="🚀")
 
-# ✅ Initialize Session State Variables
+# ✅ Initialize Session State Variables (Ensuring Chat History Persists)
 if "chat_history" not in st.session_state:
     st.session_state.chat_history = [{"role": "assistant", "content": "Hello! How can I assist you today?"}]
 if "response_ready" not in st.session_state:
@@ -28,17 +33,15 @@ if "response_ready" not in st.session_state:
 if "follow_up" not in st.session_state:
     st.session_state.follow_up = ""
 
-# ✅ Model Configuration
-model_id = "mistralai/Mistral-7B-Instruct-v0.3"
-
-# ✅ Initialize Hugging Face Model
-def get_llm_hf_inference(model_id=model_id, max_new_tokens=512, temperature=0.7):
+# ✅ Initialize Hugging Face Model (Explicitly Set to CPU/GPU)
+def get_llm_hf_inference(model_id="mistralai/Mistral-7B-Instruct-v0.3", max_new_tokens=512, temperature=0.7):
     return HuggingFaceEndpoint(
         repo_id=model_id,
         max_new_tokens=max_new_tokens,
         temperature=temperature,
         token=HF_TOKEN,
-        task="text-generation"
+        task="text-generation",
+        device=-1 if device == "cpu" else 0  # ✅ Force CPU (-1) or GPU (0)
     )
 
 # ✅ NASA API Function
@@ -50,10 +53,11 @@ def get_nasa_apod():
         return data.get("url", ""), data.get("title", ""), data.get("explanation", "")
     return "", "NASA Data Unavailable", "I couldn't fetch data from NASA right now."
 
-# ✅ Sentiment Analysis
+# ✅ Sentiment Analysis (Now Uses Explicit Device)
 sentiment_analyzer = pipeline(
     "sentiment-analysis",
-    model="distilbert/distilbert-base-uncased-finetuned-sst-2-english"
+    model="distilbert/distilbert-base-uncased-finetuned-sst-2-english",
+    device=-1 if device == "cpu" else 0  # ✅ Force CPU (-1) or GPU (0)
 )
 
 def analyze_sentiment(user_text):
@@ -66,35 +70,30 @@ def predict_action(user_text):
         return "nasa_info"
     return "general_query"
 
+# ✅ Ensure English Responses
+def ensure_english(text):
+    try:
+        detected_lang = detect(text)
+        if detected_lang != "en":
+            return "⚠️ Sorry, I only respond in English. Can you rephrase your question?"
+    except:
+        return "⚠️ Language detection failed. Please ask your question again."
+    return text
+
 # ✅ Follow-Up Question Generation
 def generate_follow_up(user_text):
-    """Generates a clean follow-up question to guide the user toward related topics or next steps."""
     prompt_text = (
         f"Given the user's question: '{user_text}', generate a SHORT follow-up question "
-        "suggesting either a related topic or asking if they need further help. "
-        "Example: 'Would you like to explore quantum superposition or ask about another physics concept?' "
-        "Keep it concise and engaging."
+        "suggesting a related topic or asking if they need more details."
     )
-    
+
     hf = get_llm_hf_inference(max_new_tokens=40, temperature=0.8)
     output = hf.invoke(input=prompt_text).strip()
-
-    # ✅ Remove unnecessary characters (like backticks and misplaced formatting)
+    
     cleaned_output = re.sub(r"```|''|\"", "", output).strip()
 
-    # ✅ Fallback in case the response is empty or invalid
     return cleaned_output if cleaned_output else "Would you like to explore another related topic or ask about something else?"
 
-# ✅ Ensure English Responses
-def ensure_english(text):
-    try:
-        detected_lang = detect(text)
-        if detected_lang != "en":
-            return "⚠️ Sorry, I only respond in English. Can you rephrase your question?"
-    except:
-        return "⚠️ Language detection failed. Please ask your question again."
-    return text
-
 # ✅ Main Response Function
 def get_response(system_message, chat_history, user_text, max_new_tokens=512):
     action = predict_action(user_text)
@@ -104,49 +103,66 @@ def get_response(system_message, chat_history, user_text, max_new_tokens=512):
         nasa_url, nasa_title, nasa_explanation = get_nasa_apod()
         response = f"**{nasa_title}**\n\n{nasa_explanation}"
         follow_up = generate_follow_up(user_text)
-        chat_history.append({'role': 'user', 'content': user_text})
-        chat_history.append({'role': 'assistant', 'content': response})
-        chat_history.append({'role': 'assistant', 'content': follow_up})
+        chat_history.extend([
+            {'role': 'user', 'content': user_text},
+            {'role': 'assistant', 'content': response},
+            {'role': 'assistant', 'content': follow_up}
+        ])
         return response, follow_up, chat_history, nasa_url
 
-    # ✅ Set Up LLM Request
+    # ✅ Invoke Hugging Face Model
     hf = get_llm_hf_inference(max_new_tokens=max_new_tokens, temperature=0.9)
 
-    # ✅ Format Chat History
     filtered_history = "\n".join(f"{msg['role']}: {msg['content']}" for msg in chat_history)
 
-    # ✅ Prompt Engineering
     prompt = PromptTemplate.from_template(
         "[INST] {system_message}\n\nCurrent Conversation:\n{chat_history}\n\n"
         "User: {user_text}.\n [/INST]\n"
-        "AI: Provide a detailed explanation with depth. "
-        "Use a conversational style, starting with 'Certainly!', 'Of course!', or 'Great question!'."
+        "AI: Provide a detailed explanation. Use a conversational tone. "
         "🚨 Answer **only in English**."
         "\nHAL:"
     )
 
-    # ✅ Invoke LLM Model
     chat = prompt | hf.bind(skip_prompt=True) | StrOutputParser(output_key='content')
     response = chat.invoke(input=dict(system_message=system_message, user_text=user_text, chat_history=filtered_history))
     response = response.split("HAL:")[-1].strip() if "HAL:" in response else response.strip()
 
-    # ✅ Ensure English
     response = ensure_english(response)
 
-    # ✅ Fallback Response
     if not response:
         response = "I'm sorry, but I couldn't generate a response. Can you rephrase your question?"
 
     follow_up = generate_follow_up(user_text)
-    chat_history.append({'role': 'user', 'content': user_text})
-    chat_history.append({'role': 'assistant', 'content': response})
-    chat_history.append({'role': 'assistant', 'content': follow_up})
+
+    chat_history.extend([
+        {'role': 'user', 'content': user_text},
+        {'role': 'assistant', 'content': response},
+        {'role': 'assistant', 'content': follow_up}
+    ])
 
     return response, follow_up, chat_history, None
 
 # ✅ Streamlit UI
 st.title("🚀 HAL - NASA AI Assistant")
 
+# ✅ Justify all chatbot responses
+st.markdown("""
+    <style>
+    .user-msg, .assistant-msg {
+        padding: 10px;
+        border-radius: 10px;
+        margin-bottom: 5px;
+        width: fit-content;
+        max-width: 80%;
+        text-align: justify;
+    }
+    .user-msg { background-color: #696969; color: white; }
+    .assistant-msg { background-color: #333333; color: white; }
+    .container { display: flex; flex-direction: column; align-items: flex-start; }
+    @media (max-width: 600px) { .user-msg, .assistant-msg { font-size: 16px; max-width: 100%; } }
+    </style>
+""", unsafe_allow_html=True)
+
 # ✅ Reset Chat Button
 if st.sidebar.button("Reset Chat"):
     st.session_state.chat_history = [{"role": "assistant", "content": "Hello! How can I assist you today?"}]
@@ -157,27 +173,23 @@ if st.sidebar.button("Reset Chat"):
 user_input = st.chat_input("Type your message here...")
 
 if user_input:
-    # ✅ Ensure `get_response()` is executed BEFORE using `response`
     response, follow_up, st.session_state.chat_history, image_url = get_response(
         system_message="You are a helpful AI assistant.",
         user_text=user_input,
         chat_history=st.session_state.chat_history
     )
 
-    # ✅ Ensure `response` is not None before using it
-    if not response:
-        response = "I'm sorry, but I couldn't generate a response."
-
-    # ✅ Display chatbot response
-    st.markdown(f"<div class='assistant-msg'><strong>HAL:</strong> {response}</div>", unsafe_allow_html=True)
+    if response:
+        st.markdown(f"<div class='assistant-msg'><strong>HAL:</strong> {response}</div>", unsafe_allow_html=True)
 
-    # ✅ Handle follow-up question
     if follow_up:
-        st.session_state.chat_history.append({'role': 'assistant', 'content': follow_up})
         st.markdown(f"<div class='assistant-msg'><strong>HAL:</strong> {follow_up}</div>", unsafe_allow_html=True)
 
-    # ✅ Display NASA image if available
     if image_url:
         st.image(image_url, caption="NASA Image of the Day")
 
     st.session_state.response_ready = True
+
+if st.session_state.response_ready and st.session_state.follow_up:
+    st.markdown(f"<div class='assistant-msg'><strong>HAL:</strong> {st.session_state.follow_up}</div>", unsafe_allow_html=True)
+    st.session_state.response_ready = False