feat(libsy): let stage algorithms fall_open tier be set at runtime - #518
feat(libsy): let stage algorithms fall_open tier be set at runtime#518ryan-lempka wants to merge 1 commit into
Conversation
|
827ada7 to
90cbb0e
Compare
90cbb0e to
d9e3d8e
Compare
be1763e to
071aa53
Compare
071aa53 to
9a9fe04
Compare
|
@coberabbitai review |
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Enterprise Run ID: 📒 Files selected for processing (3)
Included review availability: Your plan provides up to 12 included reviews per hour; 11 remain after this review. WalkthroughThe router now supports persistent fall-open tier overrides. It uses the override at zero confidence, preserves decisive signal routing, and restores the picker default after the override is cleared. ChangesFall-open tier routing
Estimated code review effort: 3 (Moderate) | ~20 minutes Merge Risk: ⚪ Minimal · up to The change is localized to runtime fallback-tier state handling, and no actionable merge-blocking risk remains after normal checks and review. Poem
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
Comment |
|
Signed-off-by: Ryan Lempka <rlempka@nvidia.com>
9a9fe04 to
0dda603
Compare
Background: Today the stage algorithm has a fallback tier that comes from a config setting called
picker. When stage algorithm is uncertain of the right model to route to it falls back to the picker. This is currently set inside of the config and fixed for the life of the process.What: This PR adds in the ability to change the fallback tier at runtime. Introduces two new functions
set_fall_openandclear_fall_openso the fallback tier is now read from session state, falling back to picker when nothing is set. Nothing calls these functions yet, so stage behavior is unchanged.Why: A first step towards hierarchical algorithms. This will allow another algorithm (such as LLM Classifier) to set which tier the stage algorithm falls back to at some point during the session. Given a large number of stage algorithm turns will default to the picker we can use LLM Classifier to adjust it based on difficulty. Follow-on PRs will look enable doing this at the start of the session, and also at the start of a user-turn.
Summary by CodeRabbit