Route generate

trainer/src/inference_server/model_benchmark.py:None–None

stub.function(
    timeout=1 * 60 * 60,
    allow_concurrent_inputs=64,
    keep_warm=1,
    concurrency_limit=1,
    gpu=modal.gpu.A100(memory=40, count=1),
    secret=modal.Secret.from_name("openpip

from typing import Union
import modal
from .main import volume, read_all_files
from ..shared import merged_model_cache_dir, lora_model_cache_dir