app.py

import gradio as gr
import numpy as np
import faiss
import torch

from sentence_transformers import SentenceTransformer
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline


# =====================================
# 1. LOAD DOCUMENTS
# =====================================
def load_documents(path="documents.txt"):
    with open(path, "r", encoding="utf-8") as f:
        docs = f.readlines()
    return [doc.strip() for doc in docs if doc.strip()]

documents = load_documents()


# =====================================
# 2. LOAD EMBEDDING MODEL (HF Open Source)
# =====================================
embedding_model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")

doc_embeddings = embedding_model.encode(documents, convert_to_numpy=True)
dimension = doc_embeddings.shape[1]


# =====================================
# 3. BUILD FAISS INDEX
# =====================================
index = faiss.IndexFlatL2(dimension)
index.add(doc_embeddings)


# =====================================
# 4. LOAD OPEN-SOURCE LLM (HF)
# =====================================
MODEL_NAME = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"  # change if needed

tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)

model = AutoModelForCausalLM.from_pretrained(
    MODEL_NAME,
    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
    device_map="auto"
)

generator = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer
)


# =====================================
# 5. RETRIEVAL FUNCTION
# =====================================
def retrieve(query, top_k=3):
    query_embedding = embedding_model.encode([query], convert_to_numpy=True)
    distances, indices = index.search(query_embedding, top_k)

    retrieved_docs = [documents[i] for i in indices[0]]
    return retrieved_docs


# =====================================
# 6. GENERIC LLM CALL
# =====================================
def call_llm(prompt):
    response = generator(
        prompt,
        max_new_tokens=300,
        do_sample=True,
        temperature=0.7,
        top_p=0.9
    )

    return response[0]["generated_text"]


# =====================================
# 7. RAG PIPELINE
# =====================================
def rag_pipeline(query):
    retrieved_docs = retrieve(query)
    context = "\n".join(retrieved_docs)

    prompt = f"""
You are a helpful AI assistant.
Answer ONLY from the provided context.

Context:
{context}

Question:
{query}

Answer:
"""

    answer = call_llm(prompt)

    return answer


# =====================================
# 8. GRADIO UI
# =====================================
with gr.Blocks() as demo:
    gr.Markdown("# 🧠 Open Source RAG (HF Only)")

    query = gr.Textbox(label="Ask your question")
    output = gr.Textbox(label="Answer")

    query.submit(rag_pipeline, query, output)

demo.launch()