[BUG] Runner terminated with signal=6

[tniccum21]

Describe the bug

Able to use model via OpenCode or Exo Chat interface.
Simple queries run, up to a few turns. Large queries cause "Runner terminated with signal=6"

To Reproduce

Steps to reproduce the behavior:

1. start Exo on both machines - nix run .#exo
2. Load Qwen3.5-397B-A17B-4bit
3. Ask series of questions

Expected behavior

Even with multiple turns model should continue to answer questions.

Actual behavior

Device 2: Get "Runner Terminated with signal=6" on Mac Studio
Device 1: Get:
1d')
[ 11:45:01.8764AM | INFO ] Fast synch flag: 1
[ 11:45:03.1247AM | INFO ] hello from the runner
[ 11:45:03.1249AM | INFO ] runner created
[ 11:45:03.1251AM | INFO ] runner shutting down
[ 11:45:03.1780AM | INFO ] bye from the runner
[ 11:45:03.1785AM | INFO ] Checking runner's status
[ 11:45:03.1787AM | INFO ] Runner was found to be alive, attempting to join process
[ 11:45:03.4638AM | INFO ] Runner exited with exit code 0
[ 11:45:03.4640AM | INFO ] Runner supervisor shutting down
[ 11:45:03.4643AM | INFO ] Runner process succesfully terminated
[ 11:45:03.5658AM | INFO ] Worker plan: CreateRunner
[ 11:45:03.6722AM | INFO ] Worker plan: Shutdown
[ 11:45:03.6726AM | INFO ] Starting task Shutdown(task_id='7da5242b-a88b-4b4e-b59f-3633073d92ed' task_status=<TaskStatus.Pending: 'Pending'> instance_id='d56dd4b9-ca34-417e-a8df-c24708e4e454' runner_id='10e265ea-2566-4d93-bd45-e0d04c33fb1d')

The above repeats over and over ad infinitum, until ctrl-c

Environment

• macOS Version: 26.4
• EXO Version: latest
• Hardware:
  • Device 1: Macbook Pro M5 Max 128GB RAM
  • Device 2: Mac Studio M2 Ultra 192GB RAM
  • Additional devices:
• Interconnection:
  • Thunderbolt 4 cable between Device 1 and 2

Additional context

[Cities2000]

6 Mac Mini M4,16G, use Qwen3.5-9b-4bit is same

[tuadcs]

I experienced the same error, seems like only happen with Qwen models (Qwen3 Coder Next, 3.5-35B-A3B, 3.5-122B-A10B), the same task run fine with GLM-4.7-Flash or gpt-oss-120b