Repository navigation
feat(serve): enable reasoning loop recovery by default - #337
Merged
Merged
Conversation
#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.
|
This pull request is labeled
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
#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
edge0-infer/)edge0-infer/tools/assets/)edge0-infer/examples/)docs/)edge0-strata/How was it tested?
Checklist
ctest --preset cpu -L "L0|L1"passes locally inedge0-infer/(engine changes); CI also runs these and theL2tests not labeledlocal-onlylocal-onlyanswer suites (ctest --preset cpu -L local-only) on Apple silicon with i8mm (M2 or later), which CI cannot judgeedge0-infer/docs/testing.md)