Give the EMPTY sentinel a concrete example in the cleanup prompt - #288
Open
melnikaite wants to merge 1 commit into
Open
Give the EMPTY sentinel a concrete example in the cleanup prompt#288melnikaite wants to merge 1 commit into
melnikaite wants to merge 1 commit into
Conversation
Filler-only dictation is supposed to produce nothing, but several supported
models return the filler verbatim instead of the EMPTY sentinel, so FreeFlow
types "ähm also ja" into the focused app. On the current default prompt,
openai/gpt-oss-20b — the default cleanup model — returned EMPTY 0/8 times for
German filler while getting English and Russian right 8/8, and
openai/gpt-oss-safeguard-20b returned it 0/8 for German, 7/8 for English and
6/8 for Russian.
Add one concrete example to the existing rule instead of restating it. This
mirrors the reasoning already recorded for "Instruction preservation is
strict": models follow concrete examples far more reliably than abstract
rules. The rule keeps its "in any language" wording, so the example does not
narrow it to English.
With the example, gpt-oss-20b returns EMPTY 8/8 in German, English and
Russian, and safeguard-20b goes to 6/8 German, 8/8 English, 8/8 Russian. A
control set of ordinary sentences that merely start with a discourse particle
("so lets move the call to tomorrow", "also lass uns den termin verschieben")
was never swallowed into EMPTY in any run.
Bump defaultSystemPromptDate so users with customized prompts see the
"default has changed" indicator.
Measured through OpenRouter rather than against Groq directly, at
temperature 0, using this prompt verbatim and the same request envelope
PostProcessingService builds.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Contributor
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (1)
📝 WalkthroughWalkthroughThe default cleanup prompt now documents filler-only transcripts with an ChangesCleanup prompt
Estimated code review effort: 1 (Trivial) | ~2 minutes Possibly related PRs
Suggested reviewers: 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
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.
Filler-only dictation is supposed to produce nothing, but several supported models return the filler verbatim instead of the EMPTY sentinel, so FreeFlow types "ähm also ja" into the focused app. On the current default prompt, openai/gpt-oss-20b — the default cleanup model — returned EMPTY 0/8 times for German filler while getting English and Russian right 8/8, and openai/gpt-oss-safeguard-20b returned it 0/8 for German, 7/8 for English and 6/8 for Russian.
Add one concrete example to the existing rule instead of restating it. This mirrors the reasoning already recorded for "Instruction preservation is strict": models follow concrete examples far more reliably than abstract rules. The rule keeps its "in any language" wording, so the example does not narrow it to English.
With the example, gpt-oss-20b returns EMPTY 8/8 in German, English and Russian, and safeguard-20b goes to 6/8 German, 8/8 English, 8/8 Russian. A control set of ordinary sentences that merely start with a discourse particle ("so lets move the call to tomorrow", "also lass uns den termin verschieben") was never swallowed into EMPTY in any run.
Bump defaultSystemPromptDate so users with customized prompts see the "default has changed" indicator.
Measured through OpenRouter rather than against Groq directly, at temperature 0, using this prompt verbatim and the same request envelope PostProcessingService builds.
Summary by CodeRabbit