Update app.py

app.py

@@ -1,9 +1,9 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
 import requests
-from bs4 import BeautifulSoup
-from bs4.element import Comment
+from bs4 import BeautifulSoup, Comment
+from llama_cpp import Llama
 
+# Function to extract visible text from a webpage
 def tag_visible(element):
     if element.parent.name in ['style', 'script', 'head', 'title', 'meta', '[document]']:
         return False
@@ -12,48 +12,49 @@ def tag_visible(element):
     return True
 
 def get_text_from_url(url):
-    response = requests.get(url)
+    response = requests.get(url, timeout=10)
     soup = BeautifulSoup(response.text, 'html.parser')
     texts = soup.find_all(text=True)
     visible_texts = filter(tag_visible, texts)
-    return "\n".join(t.strip() for t in visible_texts)
+    return " ".join(t.strip() for t in visible_texts)
 
-# Pre-fetch and truncate homepage text to reduce the prompt length.
+# Pre-fetch and truncate homepage text
 text_list = []
 homepage_url = "https://sites.google.com/view/abhilashnandy/home/"
 extensions = ["", "pmrf-profile-page"]
+
 for ext in extensions:
-    full_text = get_text_from_url(homepage_url + ext)
-    truncated_text = full_text[:1000]  # using only the first 1000 characters per extension
-    text_list.append(truncated_text)
+    try:
+        full_text = get_text_from_url(homepage_url + ext)
+        truncated_text = full_text[:2000]  # Using first 2000 characters for more context
+        text_list.append(truncated_text)
+    except Exception as e:
+        text_list.append(f"Error fetching {homepage_url+ext}: {str(e)}")
 
-SYSTEM_MESSAGE = (
-    "You are a QA chatbot to answer queries (in less than 30 words) on my homepage. "
-    "Context: " + " ".join(text_list)
-)
+CONTEXT = " ".join(text_list)
 
-# Create a Hugging Face Inference client using a CPU-friendly model.
-# Here we use 'google/flan-t5-base' as an example; you can adjust the model if needed.
-client = InferenceClient(model="google/flan-t5-base")
+# Load the Mistral model (low-latency, CPU optimized)
+llm = Llama(model_path="mistral-7b-instruct-v0.1.Q4_K_M.gguf", n_ctx=4096, n_threads=6, verbose=False)
 
+# Function to answer queries
 def answer_query(query):
-    # Compose a prompt using the system message, user query, and a reminder for a short answer.
-    prompt = SYSTEM_MESSAGE + "\nUser: " + query + "\nAnswer in less than 30 words:"
-    # Generate answer with a limit on new tokens to ensure brevity.
-    result = client.text_generation(prompt, max_new_tokens=60)
-    # Handle both list or direct string responses from the inference client.
-    if isinstance(result, list):
-        answer = result[0].get("generated_text", "")
-    else:
-        answer = result
-    return answer.strip()
-
+    prompt = (
+        "You are an AI chatbot answering queries based on the homepage of Abhilash Nandy. "
+        "Your responses should be concise (under 30 words) and directly relevant to the provided context.\n\n"
+        f"Context: {CONTEXT}\n\nUser: {query}\nAI:"
+    )
+    
+    response = llm(prompt, max_tokens=50, stop=["\nUser:", "\nAI:"], echo=False)
+    
+    return response["choices"][0]["text"].strip()
+
+# Gradio Interface
 iface = gr.Interface(
     fn=answer_query,
-    inputs=gr.Textbox(lines=2, placeholder="Enter your question here..."),
+    inputs=gr.Textbox(lines=2, placeholder="Ask a question about Abhilash Nandy..."),
     outputs="text",
     title="Homepage QA Chatbot",
-    description="A chatbot answering queries about the homepage using pre-fetched context."
+    description="Ask me anything about Abhilash Nandy's homepage."
 )
 
 if __name__ == '__main__':