app.py

import os
import zipfile
import torch  # ✅ Import torch so empty_cache works
import gradio as gr

from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.vectorstores import FAISS
from langchain.llms import HuggingFacePipeline
from langchain.chains import ConversationalRetrievalChain
from langchain.memory import ConversationBufferMemory
from langchain.prompts import PromptTemplate

# --- Step 1: Unzip FAISS index ---
if not os.path.exists("faiss_index") and os.path.exists("faiss_index.zip"):
    with zipfile.ZipFile("faiss_index.zip", "r") as zip_ref:
        zip_ref.extractall(".")

# --- Step 2: Load embedding and vectorstore ---
embedding_model = HuggingFaceEmbeddings(model_name='sentence-transformers/all-MiniLM-L6-v2')
vectordb = FAISS.load_local("faiss_index", embedding_model, allow_dangerous_deserialization=True)

# --- Step 3: Load the LLM ---
model_id = "tiiuae/falcon3-1b-instruct"
tokenizer = AutoTokenizer.from_pretrained(model_id)

# ✅ Use device_map + float16 to save memory
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    torch_dtype=torch.float16
)

pipe = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    pad_token_id=tokenizer.eos_token_id,
    max_new_tokens=200,
    do_sample=True,
    temperature=1.0,
)
llm = HuggingFacePipeline(pipeline=pipe)

# --- Step 4: Setup memory and QA chain ---
memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)

prompt = PromptTemplate.from_template("""
You are a helpful assistant at the University of Hertfordshire. Use the context below to answer the question clearly and factually.
If the answer is not in the context, say you don't know.
Context:
{context}
Question:
{question}
Helpful Answer:
""")

qa_chain = ConversationalRetrievalChain.from_llm(
    llm=llm,
    retriever=vectordb.as_retriever(search_kwargs={"k": 3}),
    memory=memory,
    chain_type="stuff",
    combine_docs_chain_kwargs={"prompt": prompt}
)

UH_LOGO = "images/UH.png"

# --- Step 5: Define chatbot logic ---
def chat(message, history):
    result = qa_chain.invoke({"question": message})
    response = result.get("answer", "")
    response = response.split("Answer:")[-1].replace("<|assistant|>", "").strip()

    # ✅ Actually clear unused GPU memory
    if torch.cuda.is_available():
        torch.cuda.empty_cache()

    return response

# --- Step 6: UI ---
sample_questions = [
    "How do I register as a new student?",
    "Where can I find accommodation?",
    "Can I renew my tenancy agreement?",
    "What do I do on my first day?",
]

with gr.Blocks() as demo:
    gr.Image(UH_LOGO, show_label=False, container=False, scale=1)
    gr.Markdown("## ASK Herts Students Help Chatbot 🤖")

    chatbot = gr.Chatbot()
    txt = gr.Textbox(placeholder="Ask me anything about university life...", label="Your question")
    submit = gr.Button("Submit")

    gr.Markdown("#### 💡 Sample Questions:")
    with gr.Row():
        for q in sample_questions:
            gr.Button(q).click(lambda x=q: gr.update(value=x), outputs=[txt])

    def respond(message, history):
        answer = chat(message, history)
        history.append((message, answer))
        return "", history

    submit.click(respond, [txt, chatbot], [txt, chatbot])
    txt.submit(respond, [txt, chatbot], [txt, chatbot])

demo.launch()