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

#16076: Adding 3-ary PriorityQueue to speed up TopFieldCollector

4 minutes ago 0 comments 0 votes 0 watches kkewwei

Description In 15140, we introduced TernaryLongHeap to optimize performance for collectors such as ... Nevertheless, TopFieldLeafCollector still adopts a 2-ary PriorityQueue, which causes identical ...

#15982 PR: Fix undercounting of RAM used by vectors buffered in in-memory segments

50 minutes ago 16 comments 0 votes 0 watches github-actions[bot], iprithv, mikemccand, shubhamvishu

Description Vector RAM accounting in ramBytesUsed() had three bugs causing IndexWriter to ... Bugs Fixed Fixes 15901 **1. BufferingKnnVectorsWriter hardcoded Float.BYTES for all encodings** Byte vectors ( ...

iprithv 50 minutes ago:	@mikemccand just wanted to touch base with you on this, in case it got buried. Thanks!
iprithv 11.3 days ago:	> Would it be better if instead of (normal + quantization overhead) we see it like keeping things ... Very likely I might be confused with this accounting code part(lemme know if thats the case) but it ... Right now we don't use FieldWriter for quantized fields as you mentioned but we do call ...

#15993 PR: Add maxValueCount to DocValuesSkipper metadata

52.7 minutes ago 7 comments 0 votes 0 watches github-actions[bot], iprithv, jainankitk, jpountz, msfroh, romseygeek

# Description resolves https://github.com/apache/lucene/issues/15794 - Add DocValuesSkippermaxV ... . - Persist exact global maxValueCount metadata in Lucene90 doc values skipper metadata for new ...

iprithv 52.7 minutes ago:	@jpountz would you be able to take a look at this? thanks!
msfroh 3.2 days ago:	> For fields that do have a skipper, this kills the weird hack where we unwrap SortedNumericDocValue ... Right! From an understandability standpoint, I think this is a big win.

#16075 PR: Fix flaky tests in TestColumnBatchIndexing

1.3 hours ago 1 comments 0 votes 0 watches Tim-Brooks, github-actions[bot]

newIndexWriterConfig() randomizes maxBufferedDocs. When multiple writes (addBatch or addDocument) exceed the flush threshold, an auto-flush occurs ... Fix all eight affected tests by calling forceMerge(1) before opening the reader.

Tim-Brooks 1.3 hours ago:

I cannot add the label, but I think this does not require a change log.

#16068 PR: FaissKnnVectorsFormat byte vectors support

1.6 hours ago 1 comments 0 votes 0 watches github-actions[bot], iprithv, kaivalnp

Implemented https://github.com/apache/lucene/issues/16067 added support for VectorSimilarityFunct ...

iprithv 1.6 hours ago:

@kaivalnp would you be able to take a look at this? thanks!

#16065 PR: FaissKnnVectorsFormat COSINE similarity support

1.7 hours ago 7 comments 0 votes 0 watches github-actions[bot], iprithv, kaivalnp, msfroh

implemented https://github.com/apache/lucene/issues/16064 read this to implement, which was ... https://github.com/facebookresearch/faiss/wiki/FAQ#how-can-i-index-vectors-for-cosine-distance

iprithv 1.4 days ago:	sure, I’ve removed faiss_Index_metric_type and on the mismatch, we should be fine now since we’re ... for perf, I ran now knnGraphTester (same as knnPerfTest) on ~10k cohere vectors \| format \| ... Thanks!
kaivalnp 2.4 days ago:	LGTM on first glance, thanks! Curious if you were able to run knnPerfTest to compare against native Lucene KNN? Also: re-normalizing every document and query vector sounds expensive?

#16069 PR: Load dense filters into bitset in MaxScoreBulkScorer

1.7 hours ago 5 comments 0 votes 0 watches github-actions[bot], iprithv, romseygeek

when the filter is dense enough (cost >= maxDoc / 32) and has a fast intoBitSet, we just load it ... one important thing I noticed is not all iterators are fast here. the default intoBitSet just loops ... also made sure wrappers don’t hide this. added delegation so we can still detect fast intoBitSet ...

iprithv 1.5 days ago:	> Thanks @iprithv, this looks like a great start! I think it's worth running the luceneutil benchmarks on this change to try and measure the effects ... sure, ran and updated in PR desc.
iprithv 1.5 days ago:	yes. makes sense, we don't need extra two bitsets. replaced with a single FilterBits and lazy ... Thanks!

«
1
2
3
4
5
»

2,563 issues

[21.1 msec search, 22.0 msec total]