Server dying with "[METAL] Command buffer execution failed: Caused GPU Timeout Error "

For the last 2 weeks or so the server has been dying with the message "Server:failed - Server exited with code 6. Auto-restart failed after 3 attempts". Looking at the server log I'm seeing the following: omlx.server - INFO - Chat completion: 5029 tokens in 247.96s (20.3 tok/s), prompt: 18729, finish_reason=stop, max_tokens=8192, request_max_tokens=None libc++abi: terminating due to uncaught exception of type std::runtime_error: [METAL] Command buffer execution failed: Caused GPU Timeout Error (00000002:kIOGPUCommandBufferCallbackErrorTimeout).

This is on an M1 Studio with 32GB ram running Qwen3.5-9B-oQ5-fp16-mtp. I'm using this as a harness for Hermes Agent, running a few cron jobs and some local LAN management tasks. When the server dies, I restart and things are fine for a day or so and then it dies again with the same message as above. I'm using the menu bar version of oMLX (0.4.4) and Mac OS Tahoe 26.5.1. I tried playing around with context size 65k (minimum required by Hermes), and 131k but the same thing happens.

I'm not an LLM expert by any means, just trying to use oMLX to experiment with local models and agents. If you need any more info let me know.