Add multilingual-e5-base as an optional model, served over HTTP - #34
Merged
Merged
Conversation
Adds an E5 option to the model list without loading another model in-process: vectors come from an external embedding service (intfloat/multilingual-e5-base, 768-d, L2-normalised) over HTTP. Both sides of an item-to-item match get the "query: " prefix, the E5 recommendation for symmetric tasks. - constants: DELOSIS_E5_BASE, appended to ALL_HARMONY_API_MODELS - settings: HARMONY_E5_EMBEDDER_URL (+ optional bearer token) - services/delosis_e5_embeddings: batched, retrying HTTP client - helpers: availability (only when the URL is configured, so stock deployments list it as unavailable and are otherwise unchanged) and vectorisation dispatch MHC and catalogue matching stay MiniLM-only exactly as before; the vectors cache is already keyed by (framework, model). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds
intfloat/multilingual-e5-baseas an optional matching model, without loading another model in-process: vectors come from an external embedding service over HTTP (768-d, L2-normalised). Stock deployments are unchanged. UnlessHARMONY_E5_EMBEDDER_URLis set, the model is listed but reported as unavailable.constants:DELOSIS_E5_BASE, appended toALL_HARMONY_API_MODELSsettings:HARMONY_E5_EMBEDDER_URL, plus an optionalHARMONY_E5_EMBEDDER_TOKENsent as a bearer tokenservices/delosis_e5_embeddings: batched, retrying HTTP clienthelpers: availability check and vectorisation dispatchMHC and catalogue matching stay MiniLM-only, exactly as before. The vectors cache is already keyed by (framework, model), so E5 and MiniLM vectors never mix.
The embedding service it calls
POST {HARMONY_E5_EMBEDDER_URL}/text/embedwith{"texts": [...], "prefix": ""}, returning{"vectors": [[...], ...]}(or"embeddings"), one vector per text. The client adds E5'squery:prefix itself, to both sides of each item-to-item match, which is E5's recommendation for symmetric tasks.prefix: ""tells the service not to add one of its own.E5's cosine similarities sit in a much narrower, higher band than MiniLM's. For example, it scores nervous / anxious at about 0.96, but also nervous / trouble sleeping at about 0.89. At the usual 70% match threshold, almost every pair matches. If E5 is offered in a frontend, it needs a higher default threshold, or the scores rescaled, before it's a sensible choice.
Independent of the other PRs.
🤖 Generated with Claude Code