Lucene: Status=Open

[Save this search]

Author relation

Updated ago

Comment count

Reaction count

Review Requested

User

Last comment user

Type

Filters: Status (Open), Issue type, Author relation, Created, Updated, Updated ago, Comment count, Reaction count, Review Requested, Mentioned, Reviewed, Commented, User, Last comment user, Draft, Component, Type, Labels, Commits?, Reporter, Assignee

#16705 PR: Prefetch full-precision vectors when rescoring

6.2 minutes ago 23 comments 0 votes 0 watches Pulkitg64, github-actions[bot], goankur, jimczi, navneet1v

Prefetch full-precision vectors when rescoring **(Implemented using AI with human in the loop)* ... Follow-up to the discussion in 16656, which this replaces. Thanks @jimczi for the review that redirected it onto the existing IndexInputprefetch hook.

Pulkitg64 6.2 minutes ago:	Hi @goankur #16628 just got merged which adds prefetch and rescoring capability for FP16 vectors. I think we need to rebase this PR with the latest changes now.
jimczi 11.9 hours ago:	LGTM, thanks. I will merge soon.

#16667 PR: Rewrite BooleanQuery to MatchNoDocsQuery when it has MUST_NOT clause

24.3 minutes ago 6 comments 0 votes 0 watches epotyom, github-actions[bot], msokolov

Rewrite BooleanQuery to MatchNoDocsQuery when it has MUST_NOT clause matching one of its MUST boost ... It's a follow-up to https://github.com/apache/lucene/pull/16446

msokolov 24.3 minutes ago:	It looks like there is a merge conflict in CHANGES -- I tried to resolve, but then it makes it look ...
epotyom 9 days ago:	Good idea, I've merged with the MUST check above. As for FILTER, if I understand correctly there is no need to unwrap boost or constant score query ...

#16734 PR: Add bulk vector path to columnar addBatch

55.9 minutes ago 0 comments 0 votes 0 watches github-actions[bot], mayya-sharipova

Builds on 16708, which added validation. Vector columns declared DENSE now go through a bulk path, and those checks come along with it via a ... Column API: - Add VectorValuesCursor<T>, a values cursor for DENSE columns modeled on ...

#16728 PR: HNSW - don't add duplicates when repairing graph links

56.5 minutes ago 8 comments 0 votes 0 watches Pulkitg64, github-actions[bot], mayya-sharipova, thecoop

There's a very small chance that a vector scorer could return that the scoring a node against ... This could result in duplicate nodes being added to a neighbor list. So check for this case before adding nodes.

Pulkitg64 56.6 minutes ago:	Do we need both i.e. for self connection and existing neighbors? ` if (neighbors.nodes()[j] == node \|\| neighbors.nodes()[j] == cnode) continue; `
mayya-sharipova 4.9 hours ago:	I find it difficult to follow labels, may be this is better: ` int cNode = candidates.nodes()[i ...

#16649: Tighten left-bound of MemoryAccountingBitsetCollectorManager.Result#bitSet

1.4 hours ago 2 comments 0 votes 0 watches Sasilekha, gaobinlong, javanna

DescriptionFollow up to #16452,which tightened the right bound of MemoryAccountingBitsetCollectorMa ... On large indices this is dominant remaining source of overallocation Suggested by @gaobinlong in ( ... Deferred from that PR because it involves a small API shape change to Result rather than a pure bug ...

Sasilekha 1.4 hours ago:	Friendly ping @javanna @gaobinlong — would appreciate design input on the Result-record shape ...
Sasilekha 21.1 days ago:	@javanna would appreciate your take on the proposed shape before I open a PR Is Result( ... Thanks!

#16663 PR: Allocate MemoryAccountingBitsetCollector bitset lazily on first match

1.4 hours ago 1 comments 0 votes 0 watches Sasilekha, gaobinlong, github-actions[bot], javanna

MemoryAccountingBitsetCollector used to reserve a bitset large enough for the entire leaf upfront, ... Under intra segment concurrency (where each partition worker gets its own collector), unmatched ... This PR delays the allocation until the first match, so collectors that see no matches allocate ...

Sasilekha 1.4 hours ago:

Gentle ping @javanna @gaobinlong this is the follow-up to #16452. Would appreciate a look when you have a moment. Happy to iterate on anything.

#16715 PR: Use unbounded lead cost for prohibited clauses

2.2 hours ago 10 comments 0 votes 0 watches github-actions[bot], kkewwei, sgup432

Description BooleanScorerSupplierbulkScorer used the positive scorer's cost as the lead cost for ... For an IndexOrDocValuesQuery , this could incorrectly select the doc-values implementation even ... This change uses an unbounded lead cost for prohibited clauses, allowing IndexOrDocValuesQuery to ...

sgup432 2.2 hours ago:	Overall LGTM. Some minor comments.
sgup432 2.2 hours ago:	Now with the recent change, maybe change this to something like "Optimize MUST_NOT scoring based on ...

«
1
2
3
4
5
»

2,666 issues

[15.7 msec search, 16.6 msec total]