Repository navigation
feat(release): publish calibrated abstention, a stack comparison, and a new drawer in v6.4.0 - #218
Merged
Merged
Conversation
… a new drawer in v6.4.0 Release highlights: - get nothing instead of noise for questions your sources cannot answer: on three real bases, 58 of 90 off-topic and 24 of 90 near-domain questions are refused, up from 41 and 1, with every answer still in place - trust weak-evidence flags again: correct answers flagged as weak fall from 43 to 14 - see how Ragmir compares: on its own repository the answering file reaches the top 3 for 76% of developer questions, against 56% for a LangChain.js pipeline and 70% for MiniSearch BM25 - reach GitHub and npm from the landing hero, and every page from a sectioned mobile drawer Release details: - **Retrieval:** each Transformers index stores a noise floor measured from 24 off-topic probe questions; only passages nearer than it count as strong vector matches, a search whose first five results are all weak answers nothing, and mxbai flags weak evidence below 30% of the question's words instead of 50% - **Diagnostics:** retired configs throw CONFIG_MIGRATION_REQUIRED, rgr doctor asks for storage compaction on fragmented indexes and for an ingest when no floor is stored, and searches no longer print LanceDB deprecation warnings - **Benchmarks:** pnpm bench:competitors compares Ragmir with LangChain.js, Orama, and MiniSearch with identical files, questions, embedding model, and vectors - **Code health:** Knip gates validation, duplicated percentile, JSON Lines, run-state, and error page code now has one implementation, and dead code is removed - **Landing:** GitHub and npm hero buttons, a drawer of card rows with Documentation and Project sections, and one shared list of project links across the hero, drawer, and footer - **Documentation:** configuration, migration, API, CLI, agent integration, troubleshooting, the agent skill, llms.txt, contributor guides, and both READMEs Verification: - pass pnpm validate with Biome, Knip, 556 Core tests and 27 landing tests with coverage gates, API, CLI, MCP and public-surface smoke checks, builds, publint, and release checks - pass the pull request checks: Quality gate, Linux x64, macOS ARM64 and Windows x64 portability on Node 22, Commitlint, and CodeQL TypeScript analysis - measure 6.3.1 against this release on three real bases with 260 answerable, 90 off-topic, and 90 near-domain questions: no false refusal and unchanged warm search latency - verify the staging landing in English and French at 375 px and 1440 px in a browser
|
🎉 This PR is included in version 6.4.0 🎉 The release is available on:
Your semantic-release bot 📦🚀 |
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.
Promotes the validated develop tree (merge of #217) to main for v6.4.0.
Release highlights:
58 of 90 off-topic and 24 of 90 near-domain questions are refused, up from 41 and 1, with
every answer still in place
developer questions, against 56% for a LangChain.js pipeline and 70% for MiniSearch BM25
Release details:
questions; only passages nearer than it count as strong vector matches, a search whose first
five results are all weak answers nothing, and mxbai flags weak evidence below 30% of the
question's words instead of 50%
compaction on fragmented indexes and for an ingest when no floor is stored, and searches no
longer print LanceDB deprecation warnings
MiniSearch with identical files, questions, embedding model, and vectors
page code now has one implementation, and dead code is removed
sections, and one shared list of project links across the hero, drawer, and footer
agent skill, llms.txt, contributor guides, and both READMEs
Verification:
API, CLI, MCP and public-surface smoke checks, builds, publint, and release checks
portability on Node 22, Commitlint, and CodeQL TypeScript analysis
90 near-domain questions: no false refusal and unchanged warm search latency