Benjamin Trent
Benjamin Trent
OK, I ran this patch against minst, sift, glove, and deep image. Recall is the exact same in all cases, so no mistakes there. There are indeed slight differences in...
@jtibshirani or @msokolov care to review? The bug was introduced back in https://github.com/apache/lucene/pull/1054
@jtibshirani My goal here was to fix the bug with as much as the original design as possible. I didn't want to spend a bunch of time re-factoring this code....
@jtibshirani I confirmed that on my M1 Macbook, there is no significant change in QPS, I tested glove-100-angular and deep-image-96-angular.
Heya @V1NAY8, so, the leaf structure in Elasticsearch machine learning has a field called: ``` number_samples ``` Having this accurately set allows for inference to supply feature importance via Shapley....
@kylemclaren RoBERTa models are not yet supported. Only Bert or DistilBert in version 8.0.0. In version 8.1.0, we will add support for MPNet. RoBERTa is on our todo list.
I, personally, love this idea. It would be a killer feature for tighter integration with Eland & Elasticsearch. I am not sure where it stands on priorities currently.
@jpountz @zhaih Looking at @dweiss old proposal I am thinking we should add a `TaskExecutor` to `IndexWriter`. The only thing thats weird is that with `ConcurrentMergeScheduler`, we already have one...
Hey @sebastian-montero this is good stuff. Take a look here: https://github.com/elastic/elasticsearch/pull/95257 In 8.8, elastic will support > 1024 dims for vector storage.
@elasticmachine update branch