Skip to content

fix(eval): glm-5.3 rejects thinking.type=disabled — send reasoning_effort alone - #62

Merged
ronaldtse merged 1 commit into
mainfrom
fix/glm-effort-payload
Aug 30, 2026
Merged

fix(eval): glm-5.3 rejects thinking.type=disabled — send reasoning_effort alone#62
ronaldtse merged 1 commit into
mainfrom
fix/glm-effort-payload

Conversation

@ronaldtse

Copy link
Copy Markdown
Contributor

No description provided.

…fort alone

Probed the live API (2026-08-31): glm-5.3-flash answers the combined
payload with HTTP 400 code 1210 — 'This model always engages in
thinking and cannot be disabled; please use low, high, or max'. The
both-knobs approach from the previous commit would have 400'd every
call and, after retries, silently scored empty predictions. Valid
reasoning_effort values are exactly low, high, max; an effort now
sends reasoning_effort only, glm-5.2 keeps thinking.type=disabled.
@ronaldtse
ronaldtse merged commit f62bed0 into main Aug 30, 2026
3 of 9 checks passed
ronaldtse added a commit that referenced this pull request Aug 31, 2026
Exhausted API retries checkpoint as empty-string predictions. Resuming
treated them as completed rows, silently scoring them as all-error —
the 2026-08-31 GLM-5.3-Flash run first read 15.96 DER because 140
such rows (produced by the pre-#62 400-ing payload) were resumed as
'done'; the true number was 8.57. Empty rows are now dropped at load
and re-fetched, with a startup line disclosing how many.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant