Cannot suppress Vulkan info messages

[dfremont]

System:

2026-06-10 09:09:56.282 Python[12015:163334862] ApplePersistenceIgnoreState: Existing state will not be touched. New state will be written to /var/folders/gw/z6d8bwbs7j914fw73vsgf3q00000gn/T/org.python.python.savedState
Inserted device layer "VK_LAYER_KHRONOS_validation" (/usr/local/share/vulkan/explicit_layer.d/../../../lib/libVkLayer_khronos_validation.dylib)
vkCreateDevice layer callstack setup to:
   <Application>
     ||
   <Loader>
     ||
   VK_LAYER_KHRONOS_validation
           Type: Explicit
           Manifest: /usr/local/share/vulkan/explicit_layer.d/VkLayer_khronos_validation.json
           Library:  /usr/local/share/vulkan/explicit_layer.d/../../../lib/libVkLayer_khronos_validation.dylib
     ||
   <Device>
       Using "Apple M4" with driver: "/usr/local/share/vulkan/icd.d/../../../lib/libMoltenVK.dylib"
Vulkan semaphores using MTLEvent.
Created VkDevice to run on GPU Apple M4 with the following 4 Vulkan extensions enabled:
		VK_KHR_portability_subset v1
		VK_KHR_shader_non_semantic_info v1
		VK_KHR_swapchain v70
		VK_EXT_extended_dynamic_state v1
Destroyed VkDevice on GPU Apple M4 with 4 Vulkan extensions enabled.
Python: 3.14.0
Platform: macOS-15.7.2-arm64-arm-64bit-Mach-O
sapien: 3.0.3
GPU: Apple M4
  Supported: 1
  Present:   1
  cudaId:    -1
  rayTrace:  0
  cudaMode   -1

Describe the bug

SAPIEN prints many Vulkan debugging messages such as the "Inserted device layer" one above, and I'm unable to suppress them. Calling sapien.render.set_log_level("off") right after importing sapien in my Python script doesn't help.

It's possible this issue isn't caused by SAPIEN itself, but I can't find any other culprits. I think the messages originate from MoltenVK, but setting the environment variable MVK_CONFIG_LOG_LEVEL=0 doesn't stop them, so maybe SAPIEN or some other code is registering a debugging callback and printing the messages itself?

To Reproduce

1. Run sapien info --all to see one example.
2. Alternatively, run a simulation in ManiSkill (this is what I'm actually doing).

Expected behavior

Calling sapien.render.set_log_level("off") should suppress these messages (or there should be some other way to suppress them).