Repository navigation
Conversation
…heaper-inference#8074) resolveEmbeddingSource() was always returning dimensions=null for remote sources, even when the model is well-known in the registry (e.g. nvidia/nv-embedqa-e5-v5 → 1024, openai/text-embedding-3-small → 1536). This caused the EmbeddingResolution.dimensions field to be unusable for callers that need to pre-allocate vector stores or detect dimension conflicts without performing an actual embed call. Fix: - Import getEmbeddingDimension from embeddingRegistry. - Call getEmbeddingDimension(model) in both remote branches (explicit 'remote' source and auto-detected remote). - Use the resolved dimension in the dimensions field and in the signature string so the cache-key includes the actual dim. - Include dim=N in the reason string when dim is known (regression guard). - Unknown models gracefully fall back to null (no registry entry). Tests: all 25 in memory-resolve-dimensions-8074 + memory-embedding-resolve pass; full memory suite 86/86 green.
…-inference#8072) The /models catalog already exposes effort_tiers for thinking-capable models, but the Combo Builder path was missing it entirely. Add effortTiers to ComboBuilderModelOption, propagate through addModelOption (both create and merge paths), and populate via CANONICAL_EFFORT_VALUES in all four model loops (synced, builtin, custom, fallback) when supportsThinking is resolved true.
…cheaper-inference#8059) Copilot device-code flow stores identity under providerSpecificData.githubEmail with no top-level email. The dedup logic gated the entire match behind 'if (tokenData.email)', so refreshes with connectionId but no email fell through to createProviderConnection, creating a duplicate on every refresh. Fix: match by connectionId first (before the email gate), then fall back to email-based dedup only when no connectionId match was found.
… paths (cheaper-inference#8081) Create stripPlaceholderFromContent() helper and apply it across all response output paths to prevent the internal sentinel '(prior reasoning summary unavailable)' from leaking into user-visible assistant content. Patched paths: - responseTranslator.ts (non-streaming, both Claude and OpenAI shapes) - openai-to-claude.ts (streaming text_delta) - openai-responses.ts (streaming Responses API text) - jsonToSse.ts (non-streaming JSON to SSE conversion) The helper uses substring matching (not exact-match) so model-added prefixes/wrappers are also caught.
- Add 'headroom' to SETTINGS_SUBOBJECT so EngineConfigPage handleSave() writes the sub-object instead of early-returning - Add headroomConfigSchema (minRows: int 2-10000, enabled: boolean) and register it in compressionSettingsUpdateSchema - Inject stored headroom config into buildStepOptions stepConfig so the runtime SmartCrusher honors the persisted minRows - Remove stale comment claiming structural engines have no detail store Fixes cheaper-inference#8056
Contributor
|
Caution The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased. |
The subagent fix (cheaper-inference#8056) handled the write-path (SETTINGS_SUBOBJECT + buildStepOptions) but the read-path switch in getCompressionSettings had no case for 'headroom'. The value was written to DB but silently ignored on read, so minRows never made it back to the UI. - Add HeadroomConfig interface + DEFAULT_HEADROOM_CONFIG to types - Add normalizeHeadroomConfig() with floor of 2 rows - Add case 'headroom' to the switch in getCompressionSettings
Collaborator
|
Thanks for this. I'm consolidating the #8056 headroom-minRows persistence on #8058 (RaviTharuma), which lands the same DB read-path fix ( |
Contributor
Author
|
Closing per maintainer feedback. Favoring #8058 (RaviTharuma) — focused single-issue PR. Will re-submit other fixes individually. |
This was referenced Jul 22, 2026
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.
Summary
Fixes #8056
The Headroom engine's minRows detail setting silently failed to persist. The UI accepted edits but
handleSave()early-returned forheadroombecause it wasn't inSETTINGS_SUBOBJECT.Root cause
SETTINGS_SUBOBJECTonly hadaggressiveandultra—handleSave()was a no-op for headroomcompressionSettingsUpdateSchemahad noheadroomsub-object — even if the UI POSTed, validation would rejectbuildStepOptions()didn't inject stored headroom config intostepConfig— SmartCrusher always gotDEFAULT_MIN_ROWS(8)Changes (3 files + 1 test)
**
EngineConfigPage.tsx\**:headroom: "headroom"toSETTINGS_SUBOBJECT**
compressionConfigSchemas.ts\**:headroomConfigSchema(minRowsint 2-10000,enabledboolean)compressionSettingsUpdateSchema**
strategySelector.ts\**:buildStepOptions()injectsoptions.config.headroomintostepConfigwhenstep.engine === "headroom"Test plan
5 regression tests verifying all three layers (UI map, schema, runtime injection).