perf: let midnight rose scale to zero (#86)

Author: sambarnes

modal/runner/containers/vllm_unified.py

@@ -176,7 +176,6 @@ def __init__(self):
     model_name=_midnight_rose,
     gpu=modal.gpu.H100(count=1),
     concurrent_inputs=4,
-    keep_warm=1,
     max_containers=1,
     quantization="GPTQ",
 )