server: --reasoning-effort-allow/-fallback and --reasoning-max-tokens-floor - #321
Open
professorpalmer wants to merge 1 commit into
Open
professorpalmer wants to merge 1 commit into
professorpalmer wants to merge 1 commit into
Conversation
…-floor --reasoning-effort-allow LIST: reasoning_effort values passed to the chat template; any other value (from the request's reasoning_effort or chat_template_kwargs) becomes --reasoning-effort-fallback instead of reaching a template that raises on it. Bonsai 2's template accepts xhigh/medium/low and raises on "high", which Cline, Kilo and Open WebUI send: every request failed with HTTP 500. --reasoning-max-tokens-floor N: with thinking on, a client max_tokens below N is raised to N. A thinking model spends its first thousands of tokens inside the thinking block; 256-4096 token app caps (PrismML's quickstart, SillyTavern, AnythingLLM, Continue) end the request there with no answer. --reasoning-budget still bounds the thinking. Both are off by default. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit a9c0710)
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
One of the five pieces of #285, split per bri-prism's request on #317. Single commit on current
prism(eaecb50c7), applies without conflicts. Server only, both flags off by default.What it does
--reasoning-effort-allow LIST/--reasoning-effort-fallback WORD: Bonsai 2's chat template raises onreasoning_effort: "high", which Cline, Kilo Code and Open WebUI send by default, so every request from those apps was an HTTP 500. An effort word outside the allow list becomes the fallback instead.--reasoning-max-tokens-floor N: with thinking on, a clientmax_tokensbelow the floor is raised to it. App defaults of 256-4096 output tokens (quickstart snippets, SillyTavern, AnythingLLM, Continue) otherwise end a thinking model mid-thought with no answer.Receipts (HumanEval 164, tests executed, the server's own sampling, same 262k/q8_0 serve)
reasoning_effort: "high"with a 256-token cap: 160/164 with the flags, 0/20 (HTTP 500) without.Samples and summaries: https://github.com/professorpalmer/bonsai-ada-surgery/tree/main/artifacts/eval/killy_20260927