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

login(token=os.environ.get("HF_TOKEN"))

BASE = "meta-llama/Llama-2-7b-hf"
LORA = "automorphic/LORA_20231221_040125_high_school_mathematics"

# Load tokenizer + model
tokenizer = AutoTokenizer.from_pretrained(BASE)
model = AutoModelForCausalLM.from_pretrained(BASE, device_map="auto")
model = PeftModel.from_pretrained(model, LORA)

# Define function for inference
def solve_math(prompt):
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    outputs = model.generate(**inputs, max_new_tokens=100)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

# Create simple Gradio UI
iface = gr.Interface(
    fn=solve_math,
    inputs="text",
    outputs="text",
    title="High School Math Solver (LoRA)",
    description="Type a math problem and the model will solve it step-by-step!"
)

iface.launch()