scripts/modal_inference.py

# Contents of deploy.py
import modal

#Setting up the image
app = modal.App("openlipsync").include(llm_app).include(web_app)
vllm_image = (
    modal.Image.from_registry("nvidia/cuda:12.8.0-devel-ubuntu22.04", add_python="3.11")
    .uv_pip_install(

    )
    .apt_install(
        "git", 
        "build-essential", 
        "cmake", 
        "curl", 
        "libcurl4-openssl-dev"
        )
    #.run_commands("")
    #.run_commands("")
    .entrypoint([])  # remove NVIDIA base container entrypoint

#Setting up the volumes
lipsync_ckpt_vol = modal.Volume.from_name("lipsync_checkpoints")
:

@app.function(
    image=lipsync_image,
    volumes={checkpoints: model_, results_dir: results},
    timeout=30 * MINUTES,
)
def inference()