[BUG] it cannot run in openclaw

[Cities2000]

Describe the bug

When I start a queue,exo display like this

04:41:01.7200PM | INFO ] Fast synch flag: 1
[ 04:41:03.0516PM | INFO ] hello from the runner
[ 04:41:03.0518PM | INFO ] runner created
[ 04:41:03.0521PM | INFO ] runner shutting down
[ 04:41:03.1078PM | INFO ] bye from the runner
[ 04:41:03.1082PM | INFO ] Checking runner's status
[ 04:41:03.1083PM | INFO ] Runner was found to be alive, attempting to join process
[ 04:41:03.4004PM | INFO ] Runner exited with exit code 0
[ 04:41:03.4008PM | INFO ] Runner supervisor shutting down
[ 04:41:03.4013PM | INFO ] Runner process succesfully terminated
[ 04:41:03.5025PM | INFO ] Worker plan: CreateRunner
[ 04:41:03.6076PM | INFO ] Worker plan: Shutdown
[ 04:41:03.6083PM | INFO ] Starting task Shutdown(task_id='68be3220-4eaa-49be-8ff7-6d7afaf75fe6' task_status=<TaskStatus.Pending: 'Pending'> instance_id='07bfddf0-4ef1-4441-bd26-e23c513c6a80' runner_id='f598aad9-d61b-4c2a-b6a0-43583c104a16')

model use qwen3.5-9b-4bit glh4.7-flash

6 MacMini M4 16*256

[rltakashige]

Thanks for raising this issue. While it's unlikely this is an issue specific to OpenClaw, there are a few reasons this could happen.

Do you have any logs from the other machines with more detail? (this one only tells us another node crashed)

[wangaocheng]

work fine with openclaw

[Cities2000]

Thanks for raising this issue. While it's unlikely this is an issue specific to OpenClaw, there are a few reasons this could happen.

Do you have any logs from the other machines with more detail? (this one only tells us another node crashed)

I am using six Mac Mini M4 devices with 16GB RAM and 256GB storage, networked via LAN. I have deployed GLM-4.7-Flash, Qwen3.5-9b-4bit, Qwen3.5-9b-8bit, Qwen3.5-35b-A3B-4bit, and Qwen3.5-35b-A3B-8bit. Basic chat functionality works fine, but once integrated with OpenClaw, memory usage spikes drastically on the machines, usually during the second call, leading to a system crash.
The error code I pasted is from one of the EXO logs. This issue mainly occurs with 4-bit models, and the error for 8-bit models is identical to #1759. What should I do?

[rltakashige]

Hmm - this seems like OpenClaw is sending far too large a message for your system to handle.

We can try adding some things to try to prevent you from going out of memory, but I'm quite interested in seeing all the logs to confirm this is the case.