We're building an agent using the Python SDK (

v1.2.5

) and are running into a consistent

TimeoutError

during worker startup, but only when we try to load the

MultilingualModel

for turn detection. The Problem: When we start our agent worker (

python main.py dev

), the main process crashes with a

TimeoutError

during

inference_executor.initialize()

. The logs show that the

lk_end_of_utterance_multilingual

inference runner is being initialized, but the main worker times out and kills the process after about 8-10 seconds. Here is the key part of the traceback: code Code

ERROR - worker failed
Traceback (most recent call last):
  ...
  File ".../livekit/agents/worker.py", line 402, in run
    await self._inference_executor.initialize()
  File ".../livekit/agents/ipc/supervised_proc.py", line 169, in initialize
    init_res = await asyncio.wait_for(...)
  File ".../asyncio/tasks.py", line 506, in wait_for
    async with timeouts.timeout(timeout):
  File ".../asyncio/timeouts.py", line 116, in __aexit__
    raise TimeoutError from exc_val
TimeoutError

What We've Tried: 1. Loading in If we initialize

MultilingualModel()

inside the

entrypoint

function, the worker starts up correctly, but then we get a

RuntimeError: no job context found

when the model tries to access the inference executor. 2. Loading in If we initialize

MultilingualModel()

inside the

prewarm

function, we get the

TimeoutError

described above. The

silero.VAD.load()

model, when placed in

prewarm

, loads successfully without a timeout. This suggests the issue is specific to the size or initialization requirements of the

MultilingualModel

. Our Question: What is the recommended architectural pattern for loading a slow, heavy, context-aware model like the

MultilingualModel

? Is there a way to extend the worker's startup timeout, or is there a different approach we should be taking to ensure it initializes reliably without crashing the worker? We are running on Windows 11 with Python 3.13. Thanks for any guidance you can provide!

Thanks for reaching out, Jeff. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.