Joe Bowser
c8d5cf5fef
Ref-hardware pool (Intel NUC12, 7.5 GiB RAM) times out at 40 min trying to run SmolLM2-360M-Q8 via native llama.cpp. The task hangs silently after the model fetch and never emits any test output. That pool exists for perf-baseline comparisons on slow hardware; our smoke test disables perfherder (perfherder: false) so it wasn't earning its keep there in the first place. Linux and Windows 24h2 main hardware continue to run the smoke test. Try run (green): https://treeherder.mozilla.org/jobs?repo=try&revision=efe6c82d0b2f7b39f64b755be730cf5b3491cef8 Differential Revision: https://phabricator.services.mozilla.com/D311676