5371: Composite embedders r=irevoire a=dureuill
# Pull Request
## Related issue
Fixes#5343
## What does this PR do?
- Implement [public usage](https://www.notion.so/meilisearch/Composite-embedder-usage-14a4b06b651f81859dc3df21e8cd02a0)
- Refactor the way we check if a parameter is mandatory/allowed/disallowed for a given source
- Take the "nesting context" into account for computer if a parameter is mandatory/allowed/disallowed
- Add tests checking all parameters with all sources, and made sure the results didn't change compared with v1.13
## Dumpless Upgrade
- This adds a new value for an existing parameter => compatible without change
- This adds new optional parameters => compatible without change
Co-authored-by: Louis Dureuil <louis@meilisearch.com>
5379: Bring back the changes from v1.13.2 into main r=dureuill a=Kerollmops
Co-authored-by: Kerollmops <Kerollmops@users.noreply.github.com>
Co-authored-by: Louis Dureuil <louis@meilisearch.com>
**Changes:**
The searchable database extraction is now relying on the AttributePatterns and FieldIdMapWithMetadata to match the field to extract.
Remove the SearchableExtractor trait to make the code less complex.
**Impact:**
- Document Addition/modification searchable indexing
- Document deletion searchable indexing
**Changes:**
The Documents changes now take a selector closure instead of a list of field to match the field to extract.
The seek_leaf_values_in_object function now uses a selector closure of a list of field to match the field to extract
The facet database extraction is now relying on the FilterableAttributesRule to match the field to extract.
The facet-search database extraction is now relying on the FieldIdMapWithMetadata to select the field to index.
The facet level database extraction is now relying on the FieldIdMapWithMetadata to select the field to index.
**Important:**
Because the filterable attributes are patterns now,
the fieldIdMap will only register the fields that exists in at least one document.
if a field doesn't exist in any document, it will not be registered even if it has been specified in the filterable fields.
**Impact:**
- Document Addition/modification facet indexing
- Document deletion facet indexing
**Changes:**
The transform structure is now relying on FieldIdMapWithMetadata and AttributePatterns to prepare
the obkv documents during a settings reindexing.
The InnerIndexSettingsDiff and InnerIndexSettings structs are now relying on FieldIdMapWithMetadata, FilterableAttributesRule and AttributePatterns to define the field and the databases that should be reindexed.
The faceted_fields_ids, localized_searchable_fields_ids and localized_faceted_fields_ids have been removed in favor of the FieldIdMapWithMetadata.
We are now relying on the FieldIdMapWithMetadata to retain vectors_fids from the facets and the searchables.
The searchable database computing is now relying on the FieldIdMapWithMetadata to know if a field is searchable and retrieve the locales.
The facet database computing is now relying on the FieldIdMapWithMetadata to compute the facet databases, the facet-search and retrieve the locales.
The facet level database computing is now relying on the FieldIdMapWithMetadata and the facet level database are cleared depending on the settings differences (clear_facet_levels_based_on_settings_diff).
The vector point extraction uses the FieldIdMapWithMetadata instead of FieldsIdsMapWithMetadata.
**Impact:**
- Dump import
- Settings update
5355: Support fetching the pooling method from the model configuration r=Kerollmops a=dureuill
# Pull Request
## Related issue
Fixes#5354
## What does this PR do?
- Fetches the pooling configuration from the model repository
- Use a pooling method that depends on the pooling configuration of that model.
- Allow overriding the pooling method with a new huggingFace embedder parameter `pooling`
- for backward-compatibility with Meilisearch v1.13
- for compatibility with embedders that exhibit the same behavior as Meilisearch v1.13
- Handle the default value of that new parameter
- for compatibility, when importing a db/a dump, it should be set to `forceMean`
- when (re)set from the settings for an embedder, it should be set to `useModel`
Co-authored-by: Louis Dureuil <louis@meilisearch.com>
5351: Bring back v1.13.0 changes into main r=irevoire a=Kerollmops
This PR brings back the changes made in v1.13 into the main branch.
Co-authored-by: ManyTheFish <many@meilisearch.com>
Co-authored-by: Kerollmops <clement@meilisearch.com>
Co-authored-by: Louis Dureuil <louis@meilisearch.com>
Co-authored-by: Clémentine <clementine@meilisearch.com>
Co-authored-by: meili-bors[bot] <89034592+meili-bors[bot]@users.noreply.github.com>
Co-authored-by: Tamo <tamo@meilisearch.com>
Co-authored-by: Clément Renault <clement@meilisearch.com>
5316: Fix the dumpless upgrade corruption r=dureuill a=irevoire
# Pull Request
## Related issue
Fixes https://github.com/meilisearch/meilisearch/issues/5280
## What does this PR do?
- Add a test that ensure we write the version in the index-scheduler even if we have a bug while writing the VERSION file
- Do what was described in the issue
Co-authored-by: Tamo <tamo@meilisearch.com>
5288: Improve AI logging r=dureuill a=Kerollmops
This PR fixes#5285 and brings the changes from #5233 to simplify debugging indexation and search performance issues related to AI. The following texts can be found in the logs to debug and understand performance issues:
- `embed_one: search` represents the time we spent waiting for the embedding generation, i.e., OpenAI, local HuggingFace, Ollama.
- `filtered_universe: search::universe` the time spent filtering the documents.
- ~`next_bucket: search::vector_sort` is the time spent finding the nearest neighbors (ANNs) in the vector store (arroy), locally~ was being triggered too many times.
- `indexing::vectors` is the time arroy spends indexing the new vectors for a batch.
- `documents::extract vectors` and `documents::merge vectors` to see the time spent generating and writing the embeddings.
Co-authored-by: Kerollmops <clement@meilisearch.com>
