Skip to content

feat(serve): enable reasoning loop recovery by default - #337

Merged
linyubupa merged 1 commit into
mainfrom
feat/reasoning-loop-recovery-default
Nov 22, 2026
Merged

linyubupa merged 1 commit into
mainfrom
feat/reasoning-loop-recovery-default

Conversation

@linyubupa

Copy link
Copy Markdown
Collaborator

#728 added the check as opt-in; the failure it covers - a reply whose reasoning repeats whole passages until it runs out of room - is common enough that the default should handle it. The default is now "recover": the server restarts the reply once from the same output with the low-effort instruction in place of the xhigh one, once per request.

"stop" is still available for a cheaper end, and false turns it off. The check still runs only while the parser is in the reasoning state, every 512 output tokens, so answer text is never affected.

What & why

Fixes #

Area

  • Engine, CLI or server (edge0-infer/)
  • Model asset tools (edge0-infer/tools/assets/)
  • Chat app examples (edge0-infer/examples/)
  • Model docs (docs/)
  • edge0-strata/
  • Other

How was it tested?

Checklist

  • Linked an agreed issue, or this PR is documentation-only
  • Related fixes in this area are in this pull request, one commit per fix
  • This is one of at most two open pull requests from me
  • ctest --preset cpu -L "L0|L1" passes locally in edge0-infer/ (engine changes); CI also runs these and the L2 tests not labeled local-only
  • Changes to model computation: ran the local-only answer suites (ctest --preset cpu -L local-only) on Apple silicon with i8mm (M2 or later), which CI cannot judge
  • Ran the higher test levels this change affects (see edge0-infer/docs/testing.md)
  • Added/updated tests for new behavior
  • Updated relevant docs / READMEs (incl. translations if user-facing)

#728 added the check as opt-in; the failure it covers - a reply whose reasoning
repeats whole passages until it runs out of room - is common enough that the
default should handle it. The default is now "recover": the server restarts the
reply once from the same output with the low-effort instruction in place of the
xhigh one, once per request.

"stop" is still available for a cheaper end, and false turns it off. The check
still runs only while the parser is in the reasoning state, every 512 output
tokens, so answer text is never affected.
@github-actions github-actions Bot added needs-info Title, issue link, test notes, or PR cap still missing strata edge0-strata docs Documentation and removed needs-info Title, issue link, test notes, or PR cap still missing labels Nov 22, 2026
@github-actions

Copy link
Copy Markdown

This pull request is labeled needs-info and is not in the review queue.
The intake rules are in CONTRIBUTING.md.

  • Replace the template comment under "## How was it tested?" with the commands, platform, and results you actually ran.

@linyubupa
linyubupa merged commit 24885ac into main Nov 22, 2026
3 of 4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

docs Documentation strata edge0-strata

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant