app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM
from huggingface_hub import login
import torch
import os

# Login with token from HF Space Secrets (Settings → Secrets)
login(os.environ["HF_TOKEN"])

model_id = "UBC-NLP/NileChat-3B"

# Load tokenizer and model with authentication
tokenizer = AutoTokenizer.from_pretrained(model_id, token=os.environ["HF_TOKEN"])
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto",
    token=os.environ["HF_TOKEN"]

)

# Simple prompt → reply function
def chat(prompt):
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    output = model.generate(
        **inputs,
        max_new_tokens=256,
        do_sample=True,
        top_p=0.95,
        temperature=0.8
    )
    decoded = tokenizer.decode(output[0], skip_special_tokens=True)
    return decoded.replace(prompt, "").strip()

# Gradio chat interface
gr.ChatInterface(chat, title="NileCoach 🇪🇬 Masry LLM Agent").launch()