4548: v1.8 hybrid search changes r=dureuill a=dureuill
Implements the search changes from the [usage page](https://meilisearch.notion.site/v1-8-AI-search-API-usage-135552d6e85a4a52bc7109be82aeca42#40f24df3da694428a39cc8043c9cfc64)
### ⚠️ Breaking changes in an experimental feature:
- Removed the `_semanticScore`. Use the `_rankingScore` instead.
- Removed `vector` in the response of the search (output was too big).
- Removed all the vectors from the `vectorSort` ranking score details
- target vector appearing in the name of the rule
- matched vector appearing in the details of the rule
### Other user-facing changes
- Added `semanticHitCount`, indicating how many hits were returned from the semantic search. This is especially useful in the hybrid search.
- Embed lazily: Meilisearch no longer generates an embedding when the keyword results are "good enough".
- Graceful embedding failure in hybrid search: when doing hybrid search (`semanticRatio in ]0.0, 1.0[`), an embedding failure no longer causes the search request to fail. Instead, only the keyword search is performed. When doing a full vector search (`semanticRatio==1.0`), a failure to embed will still result in failing that search.
Co-authored-by: Louis Dureuil <louis@meilisearch.com>
4549: Hugging Face embedder improvements r=dureuill a=dureuill
Architectural changes/Internal improvements
### 1. Prefer safetensors weights over pytorch weights when available
safetensors weights are memory mapped, which reduces memory usage of supported models.
### 2. Update candle
Updates candle to `0.4.1`, now targeting crates.io and the tokenizers to `v0.15.2` (still on github).
This might fix https://github.com/meilisearch/meilisearch/issues/4399 thanks to the now included https://github.com/huggingface/candle/issues/1454
Co-authored-by: Louis Dureuil <louis@meilisearch.com>
4456: Add Ollama as an embeddings provider r=dureuill a=jakobklemm
# Pull Request
## Related issue
[Related Discord Thread](https://discord.com/channels/1006923006964154428/1211977150316683305)
## What does this PR do?
- Adds Ollama as a provider of Embeddings besides HuggingFace and OpenAI under the name `ollama`
- Adds the environment variable `MEILI_OLLAMA_URL` to set the embeddings URL of an Ollama instance with a default value of `http://localhost:11434/api/embeddings` if no variable is set
- Changes some of the structs and functions in `openai.rs` to be public so that they can be shared.
- Added more error variants for Ollama specific errors
- It uses the model `nomic-embed-text` as default, but any string value is allowed, however it won't automatically check if the model actually exists or is an embedding model
Tested against Ollama version `v0.1.27` and the `nomic-embed-text` model.
## PR checklist
Please check if your PR fulfills the following requirements:
- [x] Does this PR fix an existing issue, or have you listed the changes applied in the PR description (and why they are needed)?
- [x] Have you read the contributing guidelines?
- [x] Have you made sure that the title is accurate and descriptive of the changes?
Co-authored-by: Jakob Klemm <jakob@jeykey.net>
Co-authored-by: Louis Dureuil <louis.dureuil@gmail.com>
Instead of the user manually specifying the model dimensions it will now automatically get determined
Just like with hf.rs the word "test" gets embedded to determine the dimensions of the output
Add a dedicated error type for if the model doesn't exist (don't automatically pull it though) and set the fault of that error to be the user
Initial prototype of Ollama embeddings actually working, error handlign / retries still missing.
Allow model to be any String and require dimensions parameter
Fixed rustfmt formatting issues
There were some formatting issues in the initial PR and this should not make the changes comply with the Rust style guidelines
Because I accidentally didn't follow the style guide for commits in my commit messages I squashed them into one to comply
-> make sure the settings change is rejected or the settings task fails when the specified model doesn't support
overriding `dimensions` and the passed `dimensions` differs from the model's default dimensions.
- DistributionShift in Search object (to be set from model in embed?)
- Fix issue where embedder index wasn't computed at search time
- Accept as default embedder either the "default" one, or the only embedder when there is only one
