Wasmtime GitHub notifications bot (Jul 14 2026 at 21:59): Wasmtime GitHub notifications bot (Jul 14 2026 at 21:59):alexcrichton opened issue #13880:
Today in particular, but historically as well, there have been a lot of failures in the wasmtime-wasi-nn crate. Some sample logs are:
https://github.com/bytecodealliance/wasmtime/actions/runs/29358428549/job/87172005655
https://github.com/bytecodealliance/wasmtime/actions/runs/29350702096/job/87145948188
https://github.com/bytecodealliance/wasmtime/actions/runs/29340495721/job/87110767184@jlb6740 and @rahulchaphalkar as the defacto-owners of wasmtime-wasi-nn would y'all be able to take a look at this? If this continue to be onerous we'll probably need to disable the tests as it can hold up landing PRs.
Wasmtime GitHub notifications bot (Jul 14 2026 at 21:59):alexcrichton added the ci label to Issue #13880.
Wasmtime GitHub notifications bot (Jul 14 2026 at 23:45):alexcrichton commented on issue #13880:
Wasmtime GitHub notifications bot (Jul 15 2026 at 05:16):alexcrichton commented on issue #13880:
Wasmtime GitHub notifications bot (Jul 15 2026 at 05:17):alexcrichton commented on issue #13880:
Wasmtime GitHub notifications bot (Jul 15 2026 at 23:44):rahulchaphalkar commented on issue #13880:
Thanks Alex for bringing this to our attention. I'm addressing some of these issues in https://github.com/bytecodealliance/wasmtime/pull/13886 (still a draft). Trying to repro
illegal instructionerror which seems to be on windows.
Wasmtime GitHub notifications bot (Jul 16 2026 at 15:21):alexcrichton commented on issue #13880:
Thanks for looking into this! I'm going to keep logging things here so there's a set of logs to draw from if needed as a heads up:
Wasmtime GitHub notifications bot (Jul 16 2026 at 15:54):alexcrichton commented on issue #13880:
Wasmtime GitHub notifications bot (Jul 17 2026 at 14:30):alexcrichton commented on issue #13880:
Wasmtime GitHub notifications bot (Jul 20 2026 at 14:28):alexcrichton commented on issue #13880:
Wasmtime GitHub notifications bot (Jul 20 2026 at 14:29):alexcrichton commented on issue #13880:
Wasmtime GitHub notifications bot (Jul 28 2026 at 19:10):alexcrichton commented on issue #13880:
@rahulchaphalkar this is a failure after your openvino update landed
Wasmtime GitHub notifications bot (Jul 28 2026 at 21:00):rahulchaphalkar commented on issue #13880:
Bummer. Thanks for letting me know. I have opened a bug in onednn https://github.com/uxlfoundation/oneDNN/issues/5689 , and will keep track of it. I don't have the exact Xeon family of platform available to repro it reliably for now, I will keep this issue updated.
Wasmtime GitHub notifications bot (Jul 30 2026 at 23:12):rahulchaphalkar commented on issue #13880:
After investigation, the issue turns out to be inconsistent CPUID contents on github runners for a specific CPU. This ultimately causes OneDNN to execute an AMX instruction which causes illegal instruction error.
Bug report - https://github.com/actions/runner-images/issues/14483
Illegal instruction from OneDNN - https://github.com/uxlfoundation/oneDNN/issues/5689#issuecomment-5136593460
This is now a github runner + virtualization issue, so it should occur only in CI.
@alexcrichton I can limit OneDNN ISA to not emit AMX in CI, that should suffice.
Wasmtime GitHub notifications bot (Jul 31 2026 at 01:58):alexcrichton commented on issue #13880:
Oh wow that's a nasty bug! Currently we're not gating on wasi-nn testing which means that it's allowed to fail and we can still merge things, so no urgent action is necessarily needed. Perhaps we can see if that upstream issue gets attention and maybe wait for that to get fixed? If you've got a fix to limit testing though I think that'd also be reasonable to add and would be happy to review/merge that
Last updated: Aug 30 2026 at 09:07 UTC