import gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
import torch

model_name = "ajibawa-2023/Young-Children-Storyteller-Mistral-7B"
quant_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.float16,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_quant_type="nf4"
)

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    quantization_config=quant_config,
    device_map="auto",
    torch_dtype=torch.float16,
    trust_remote_code=True
)

def generate_story(prompt, max_length=400, temperature=0.7, top_p=0.9):
    formatted_prompt = f"### Instruction:\nCreate a story for young children about: {prompt}\n\n### Response:\n"
    inputs = tokenizer.encode(formatted_prompt, return_tensors="pt").to(model.device)
    outputs = model.generate(
        inputs,
        max_length=max_length,
        temperature=temperature,
        top_p=top_p,
        do_sample=True,
        pad_token_id=tokenizer.eos_token_id,
        repetition_penalty=1.1
    )
    return tokenizer.decode(outputs[0], skip_special_tokens=True).split("### Response:")[-1].strip()

gr.Interface(fn=generate_story, inputs="text", outputs="text").launch()