import gradio as gr
import faiss
import pickle
from transformers import AutoTokenizer, AutoModelForCausalLM
from sentence_transformers import SentenceTransformer

tokenizer = AutoTokenizer.from_pretrained("tiny-gpt2-finetuned-ajem")
model = AutoModelForCausalLM.from_pretrained("tiny-gpt2-finetuned-ajem")
embedder = SentenceTransformer("all-MiniLM-L6-v2")

index = faiss.read_index("rag_index.faiss")
with open("rag_texts.pkl", "rb") as f:
    texts = pickle.load(f)

def get_context(query, top_k=3):
    q_emb = embedder.encode([query])
    D, I = index.search(q_emb, top_k)
    return "\n".join([texts[i] for i in I[0]])

def chat(query):
    context = get_context(query)
    prompt = context + "\nUsuario: " + query + "\nAsistente:"
    inputs = tokenizer(prompt, return_tensors="pt")
    outputs = model.generate(**inputs, max_new_tokens=100, pad_token_id=tokenizer.eos_token_id)
    response = tokenizer.decode(outputs[0], skip_special_tokens=True).replace(prompt, "").strip()
    return response

gr.ChatInterface(chat, title="Tiny Chat RAG").launch()