import gradio as gr
import torch
from huggingface_hub import hf_hub_download

from gradio_tabs.animation import animation
from gradio_tabs.vid_edit import vid_edit
from gradio_tabs.img_edit2 import img_edit
from networks.generator import Generator

# Optimize torch.compile performance
torch.set_float32_matmul_precision('high')  # Enable TensorFloat32 for better performance
torch._dynamo.config.cache_size_limit = 64  # Increase cache size to reduce recompilations

device = torch.device("cuda")
gen = Generator(size=512, motion_dim=40, scale=2).to(device)
ckpt_path = hf_hub_download(repo_id="YaohuiW/LIA-X", filename="lia-x.pt")
gen.load_state_dict(torch.load(ckpt_path, weights_only=True))
gen.eval()

chunk_size=30

def load_file(path):

	with open(path, 'r', encoding='utf-8') as f:
		content  = f.read()

	return content

custom_css = """
<style>
  body {
	font-family: Georgia, serif; /* Change to your desired font */
  }
  h1 {
	color: black; /* Change title color */
  }
</style>
"""

with gr.Blocks(css=custom_css, theme=gr.themes.Soft()) as demo:
	gr.HTML(load_file("assets/title.md"))
	with gr.Row():
		with gr.Accordion(open=False, label="Instruction"):
			gr.Markdown(load_file("assets/instruction.md"))
	
	with gr.Row():
		with gr.Tabs():
			animation(gen, chunk_size, device)
			img_edit(gen, device)
			vid_edit(gen, chunk_size, device)


demo.launch(allowed_paths=["./data/source","./data/driving"])