import gradio as gr
import torch
from transformers import AutoProcessor, LlavaForConditionalGeneration
from transformers import BitsAndBytesConfig

quantization_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.float16
)

model_id = "llava-hf/llava-1.5-7b-hf"

processor = AutoProcessor.from_pretrained(model_id)
model = LlavaForConditionalGeneration.from_pretrained(
    model_id,
    quantization_config=quantization_config,
    device_map="auto"
)


def text_to_image(image, prompt):
    prompt = f'USER: <image>\n{prompt}\nASSISTANT:'

    inputs = processor([prompt], images=[image], padding=True, return_tensors="pt").to(model.device)
    for k, v in inputs.items():
        print(k, v.shape)
    print(inputs)
    output = model.generate(**inputs, max_new_tokens=100)
    generated_text = processor.batch_decode(output, skip_special_tokens=True)
    for text in generated_text:
        print(text.split("ASSISTANT:")[-1])

print()
demo = gr.Interface(
    fn=text_to_image,
    inputs=[
        gr.Image(label='Select an image to analyze', type='pil'),
        gr.Textbox(label='Enter Prompt')
    ],
    outputs=gr.Textbox(label='Maurice says:')
)

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