use qdrant vectors in hybrid search #2718
Conversation
OpenAPI ChangesShow/hide No detectable change.Unexpected changes? Ensure your branch is up-to-date with |
abeglova
force-pushed
the
ab/use-qdrant-vector
branch
2 times, most recently
from
11eb260 to
2bca7f2
Compare
shanbady
left a comment
shanbady
left a comment
There was a problem hiding this comment.
Functionally it works great! just some minor cleanup suggestions/comments
vector_search/utils.py
Outdated
| } | ||
|
|
||
|
|
||
| def get_vector_for_learning_resource(readable_id): |
There was a problem hiding this comment.
Might be cleaner to add a "with_vectors" as an optional parameter to retrieve_points_matching_params and use that:
retrieve_points_matching_params({"readable_id": "course-234"}, with_vectors=True)[0]
learning_resources_search/api.py
Outdated
| } | ||
| encoder = dense_encoder() | ||
| query_vector = encoder.embed_query(text) | ||
| vector_query = {"knn": {"vector_embedding": {"vector": query_vector, "k": 5}}} |
There was a problem hiding this comment.
might be good to have "k" and "pagination_depth" defined in constants or in settings.py (or omitted altogether if they default to something reasonable)
learning_resources_search/api.py
Outdated
| "combination": { | ||
| "technique": "arithmetic_mean", | ||
| "parameters": {"weights": [0.6, 0.2, 0.2]}, | ||
| "parameters": {"weights": [0.8, 0.2]}, |
There was a problem hiding this comment.
these also seem like magic numbers we may tweak often. consider moving to settings or as a constant. since all of search_pipeline is static json we could do something like:
OPENSEARCH_HYBRID_PIPELINE_CONFIGURATION = {...}
| } | ||
| } | ||
|
|
||
| if object_type == COMBINED_INDEX: |
There was a problem hiding this comment.
imo the naming of these indexes are getting a bit confusing (constants.COMBINED_INDEX vs constants.ALL_INDEXES vs constants.BOTH_INDEXES) might be better to call this HYBRID_INDEX or HYBRID_COMBINED_INDEX etc
abeglova
force-pushed
the
ab/use-qdrant-vector
branch
from
0a3af13 to
3586b9e
Compare
What are the relevant tickets?
closes https://github.com/mitodl/hq/issues/9148
Description (What does it do?)
This pr updates the hybrid search index in opensearch to reuse vectors from qdrant. The hybrid search is currently behind a flag (search_mode=hybrid)
How can this be tested?
set
OPENAI_API_KEY to the value from rc
and
QDRANT_ENCODER=vector_search.encoders.litellm.LiteLLMEncoder
run
docker-compose run web ./manage.py generate_embeddings --skip-contentfilesto ensure you have resource embedding in qdrantrun
docker-compose run web ./manage.py recreate_index --combined_hybridGot to
http://open.odl.local:8062/search
search for "intro to ai course" you should have no results. The search with search mode not set should behave normally
select search_mode=hybrid from the admin options panel (login required) or just add it to the url
search for "intro to ai course" you should get results. Facets and base search should also still work