app.py

import gradio as gr
import os
import json
import faiss
import numpy as np
import torch
from sentence_transformers import SentenceTransformer
from huggingface_hub import InferenceClient, hf_hub_download

# 🔹 Hugging Face Credentials
HF_REPO = "Futuresony/future_ai_12_10_2024.gguf"
HF_TOKEN = os.getenv('HUGGINGFACEHUB_API_TOKEN')  # Store your token as an environment variable for security

# 🔹 FAISS Index Path
FAISS_PATH = "asa_faiss.index"

# 🔹 Load Sentence Transformer for Embeddings
embedder = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")

# 🔹 Load FAISS Index from Hugging Face
faiss_local_path = hf_hub_download(HF_REPO, "asa_faiss.index", token=HF_TOKEN)
faiss_index = faiss.read_index(faiss_local_path)

# 🔹 Initialize Hugging Face Model Client
client = InferenceClient(model=HF_REPO, token=HF_TOKEN)

# 🔹 Retrieve Relevant FAISS Context
def retrieve_relevant_context(user_query, top_k=3):
    query_embedding = embedder.encode([user_query], convert_to_tensor=True).cpu().numpy()
    distances, indices = faiss_index.search(query_embedding, top_k)

    retrieved_texts = []
    for idx in indices[0]:  # Extract top_k results
        if idx != -1:  # Ensure valid index
            retrieved_texts.append(f"Example: {idx} → {idx}")  # Customize how retrieved data appears

    return "\n".join(retrieved_texts) if retrieved_texts else "No relevant data found."

# 🔹 Format Model Prompt with FAISS Guidance
def format_prompt(user_input, system_prompt, history):
    retrieved_context = retrieve_relevant_context(user_input)

    faiss_instruction = (
        "Use the following example responses as a guide for formatting and writing style:\n"
        f"{retrieved_context}\n\n"
        "### Instruction:\n"
        f"{user_input}\n\n### Response:"
    )

    return faiss_instruction

# 🔹 Chatbot Response Function
def respond(message, history, system_message, max_tokens, temperature, top_p):
    full_prompt = format_prompt(message, system_message, history)

    response = client.text_generation(
        full_prompt,
        max_new_tokens=max_tokens,
        temperature=temperature,
        top_p=top_p,
    )

    # ✅ Extract only model-generated response
    cleaned_response = response.split("### Response:")[-1].strip()
    
    history.append((message, cleaned_response))  # ✅ Update chat history
    
    yield cleaned_response  # ✅ Output the response

# 🔹 Gradio Chat Interface
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a helpful AI trained to follow FAISS-based writing styles.", label="System message"),
        gr.Slider(minimum=1, maximum=250, value=128, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.9, step=0.1, label="Temperature"),
        gr.Slider(minimum=0.1, maximum=1.0, value=0.99, step=0.01, label="Top-p (nucleus sampling)"),
    ],
)

if __name__ == "__main__":
    demo.launch()