Skip to content

[bm25 5/5] refactor: drop in-memory BM25Scorer; FTS5 is the only implementation#335

Open
raoabinav wants to merge 5 commits into
yichuan-w:mainfrom
raoabinav:refactor/drop-in-memory-bm25
Open

[bm25 5/5] refactor: drop in-memory BM25Scorer; FTS5 is the only implementation#335
raoabinav wants to merge 5 commits into
yichuan-w:mainfrom
raoabinav:refactor/drop-in-memory-bm25

Conversation

@raoabinav
Copy link
Copy Markdown
Contributor

Sub-PR 5 of 5 from #327, the final step. Stacks on the default flip in the prior sub-PR.

Breaking change for indexes without an FTS5 artifact. Documented in the runtime error message; affects only indexes built before the FTS5 backend (or with bm25_backend="memory" during the soak window). Users will see a clear RuntimeError on first hybrid/BM25 call pointing them at leann rebuild.

Deletes BM25Scorer entirely along with the build-time pickle snapshot path (_build_bm25_snapshot, the bm25_snapshot meta.json field, the prebuild_bm25 kwarg) and the fit-on-search fallback inside _init_bm25. Drops the now-orphan from collections import Counter, defaultdict.

bm25_backend is now fts5-only; passing "memory" raises ValueError.

Net: -74 LOC of dead BM25 implementation, one canonical path.

BM25 persistence train (#327)

raoabinav and others added 4 commits May 20, 2026 11:04
Pure refactor, no behavior change. Sets up follow-up PRs that add an FTS5-
backed implementation behind the same contract.

Currently the only consumer of BM25 (LeannSearcher._init_bm25 / _bm25_search)
relies on `BM25Scorer.fit(passages)` and `BM25Scorer.search(query, top_k) ->
list[SearchResult]`. Extracting those two methods into an ABC makes the
follow-up FTS5 implementation drop-in. See yichuan-w#327 for the broader plan.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sub-PR 2 of 5 from yichuan-w#327. Builds on yichuan-w#328 (BM25Index ABC).

Adds `LeannBuilder(prebuild_bm25=True)`. When set, build_index fits a
BM25Scorer on the chunks and pickles it to <index>.bm25.pkl, then records
the snapshot filename in meta.json under "bm25_snapshot".

LeannSearcher._init_bm25 now checks for that snapshot first: if present and
loads cleanly, it skips fitting; otherwise it falls back to today's behavior
(scan passages.jsonl and fit). Older indexes are unaffected — no snapshot
field in their meta.json, so the fit-on-search path runs.

Default stays False so this PR changes nothing for existing callers. Default
flip happens in sub-PR 4 once the FTS5 backend (sub-PR 3) lands.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sub-PR 3 of 5 from yichuan-w#327. Builds on yichuan-w#328 and yichuan-w#332.

New `Fts5BM25Index(BM25Index)` class backed by SQLite FTS5 (`tokenize='unicode61 remove_diacritics 2'`). fit() bulk-inserts into a fresh virtual table; search() runs `MATCH` with `-bm25()` ordering so the rest of LeannSearcher (and hybrid fusion) keeps higher-is-better.

Opt-in via `LeannBuilder(bm25_backend="fts5")`. When set, build_index writes `<index>.bm25.sqlite` and records `bm25_backend="fts5"` + `bm25_db` in meta.json. `LeannSearcher._init_bm25` honors the field: fts5 → mmap the sqlite; memory → use the pickle from sub-PR 2; absent → fall back to fit-on-search for older indexes.

Default `bm25_backend="memory"` so nothing changes for existing callers. Default flip happens in sub-PR 4.

Query tokenization matches BM25Scorer (strip punctuation, lowercase, OR terms) so the same query text behaves consistently across backends; FTS5 syntax surprises like `:` `*` get neutralized.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sub-PR 4 of 5 from yichuan-w#327. Builds on the FTS5 backend in sub-PR 3.

Flips LeannBuilder(bm25_backend=...) default from "memory" to "fts5". New
indexes built without specifying a backend now write a sqlite FTS5 database
instead of in-memory state. Existing indexes are unaffected — their meta.json
either lacks bm25_backend (older, fit-on-search path) or has it set
explicitly.

In-memory `BM25Scorer` stays as `bm25_backend="memory"` for tiny corpora
where the FTS5 sqlite overhead isn't worth it. Sub-PR 5 removes it after
this default has soaked.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@raoabinav raoabinav force-pushed the refactor/drop-in-memory-bm25 branch from ce1ed22 to 329c903 Compare May 20, 2026 23:25
@raoabinav raoabinav changed the title refactor: drop in-memory BM25Scorer; FTS5 is the only BM25 implementation [bm25 5/5] refactor: drop in-memory BM25Scorer; FTS5 is the only implementation May 20, 2026
…tion

Sub-PR 5 of 5 from yichuan-w#327. Builds on sub-PR 4 (FTS5 default).

Deletes BM25Scorer entirely along with the build-time pickle snapshot path
(_build_bm25_snapshot, the "bm25_snapshot" meta.json field, the prebuild_bm25
kwarg) and the fit-on-search fallback inside _init_bm25. Drops the now-orphan
`from collections import Counter, defaultdict`.

`bm25_backend` is now fts5-only; passing "memory" raises ValueError. Indexes
built before this had a BM25 artifact will get a clear RuntimeError on first
hybrid/BM25 call pointing the user at `leann rebuild`.

Net: -74 LOC of dead BM25 implementation, one canonical path.

Breaking change for indexes without an FTS5 artifact. Documented in the error
message; affects only indexes built pre-sub-PR-3 (or with bm25_backend="memory"
during the soak window between sub-PRs 2-4).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@raoabinav raoabinav force-pushed the refactor/drop-in-memory-bm25 branch from 329c903 to 0100431 Compare May 21, 2026 02:24
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant