David Kyle

[ML] Write downloaded model parts async

5

It has been observed that downloading and installing the built in `.elser_model_2` and `.multilingual-e5-small` models is much slower than expected. The cause is in the `ModelImporter` class which downloads the...

Elasticsearch has upgraded to PyTorch to 2.3.1 in 8.16 and 8.15.1. This PR updates Eland to use the same version. The usual compatibility issues apply, Eland will refuse to upload...

Requires an upgrade of PyTorch to at least version 2.2.0

The BERT Japanese Tokenizer requires the following packages that are not including in the Docker image. - fugashi - ipadic - unidic_lite The list comes from the PR description where...

[CI] SearchHitsTests testFromXContent failing

7

**Build scan:** https://gradle-enterprise.elastic.co/s/a7fxyrffc2bhs/tests/:server:test/org.elasticsearch.search.SearchHitsTests/testFromXContent **Reproduction line:** ``` ./gradlew ':server:test' --tests "org.elasticsearch.search.SearchHitsTests.testFromXContent" -Dtests.seed=FD701560D64B1D14 -Dtests.locale=sr-ME -Dtests.timezone=Atlantic/Madeira -Druntime.java=20 ``` **Applicable branches:** 8.9 **Reproduces locally?:** Yes **Failure history:** https://gradle-enterprise.elastic.co/scans/tests?tests.container=org.elasticsearch.search.SearchHitsTests&tests.test=testFromXContent **Failure excerpt:** ``` java.lang.AssertionError: Expected: but:...

[NLP] Enable import of models with missing vocabulary files

1

Eland needs access to a model's vocabulary file so that is can be uploaded to Elasticsearch along with the model definition. In some cases the vocab file is not included...

Support the Jina AI jina-embeddings-v3 model

6

When trying to import the `jina-embeddings-v3` model I encountered the error described in https://huggingface.co/jinaai/jina-embeddings-v3/discussions/96. The stated fix on the discussion is to install the `flash-attn` package. Support may depend on...

Use the fast tokenizer

1

PR for testing For #721