Merge pull request #2103 from weaviate/fix_hybrid_tests

Fix hybrid tests

Author: dirkkul

integration/test_collection_diversity_hybrid.py

@@ -1,10 +1,9 @@
 """Integration tests for hybrid search + MMR diversity selection.
 
-``DiversitySelection`` passed inside ``HybridVector.near_vector`` /
-``HybridVector.near_text`` is applied by the server as a post-fusion MMR pass
-(Weaviate >= 1.39.0). These tests assert that ``balance=0`` (pure diversity)
-produces a different ordering than ``balance=1`` (pure relevance), and that
-``mmr.limit`` caps the result count.
+``diversity_selection`` is a hybrid-level argument: the server applies it as a
+post-fusion MMR pass (Weaviate >= 1.38.6). These tests assert that ``balance=0``
+(pure diversity) produces a different ordering than ``balance=1`` (pure
+relevance), and that ``mmr.limit`` caps the result count.
 
 The equivalent ``near_vector`` behaviour is covered in
 ``test_collection_diversity.py``.
@@ -68,18 +67,14 @@ def test_hybrid_near_vector_balance_0_differs_from_balance_1(
     collection = _create_clustered_collection(collection_factory)
     balance_0 = collection.query.hybrid(
         query=None,
-        vector=HybridVector.near_vector(
-            vector=[1.0, 0.0, 0.0],
-            diversity_selection=Diversity.mmr(limit=3, balance=0.0),
-        ),
+        vector=HybridVector.near_vector(vector=[1.0, 0.0, 0.0]),
+        diversity_selection=Diversity.mmr(limit=3, balance=0.0),
         limit=3,
     ).objects
     balance_1 = collection.query.hybrid(
         query=None,
-        vector=HybridVector.near_vector(
-            vector=[1.0, 0.0, 0.0],
-            diversity_selection=Diversity.mmr(limit=3, balance=1.0),
-        ),
+        vector=HybridVector.near_vector(vector=[1.0, 0.0, 0.0]),
+        diversity_selection=Diversity.mmr(limit=3, balance=1.0),
         limit=3,
     ).objects
     assert [o.uuid for o in balance_0] != [o.uuid for o in balance_1]
@@ -97,10 +92,8 @@ def test_hybrid_near_vector_balance_1_matches_baseline(
     ).objects
     mmr_balance_1 = collection.query.hybrid(
         query=None,
-        vector=HybridVector.near_vector(
-            vector=[1.0, 0.0, 0.0],
-            diversity_selection=Diversity.mmr(limit=3, balance=1.0),
-        ),
+        vector=HybridVector.near_vector(vector=[1.0, 0.0, 0.0]),
+        diversity_selection=Diversity.mmr(limit=3, balance=1.0),
         limit=3,
     ).objects
     assert [o.uuid for o in baseline] == [o.uuid for o in mmr_balance_1]
@@ -114,19 +107,15 @@ def test_hybrid_alpha_1_balance_0_differs_from_balance_1(
     balance_0 = collection.query.hybrid(
         query="irrelevant",
         alpha=1.0,
-        vector=HybridVector.near_vector(
-            vector=[1.0, 0.0, 0.0],
-            diversity_selection=Diversity.mmr(limit=3, balance=0.0),
-        ),
+        vector=HybridVector.near_vector(vector=[1.0, 0.0, 0.0]),
+        diversity_selection=Diversity.mmr(limit=3, balance=0.0),
         limit=3,
     ).objects
     balance_1 = collection.query.hybrid(
         query="irrelevant",
         alpha=1.0,
-        vector=HybridVector.near_vector(
-            vector=[1.0, 0.0, 0.0],
-            diversity_selection=Diversity.mmr(limit=3, balance=1.0),
-        ),
+        vector=HybridVector.near_vector(vector=[1.0, 0.0, 0.0]),
+        diversity_selection=Diversity.mmr(limit=3, balance=1.0),
         limit=3,
     ).objects
     assert [o.uuid for o in balance_0] != [o.uuid for o in balance_1]
@@ -141,9 +130,7 @@ def test_hybrid_respects_mmr_limit(
 
     result = collection.query.hybrid(
         query=None,
-        vector=HybridVector.near_vector(
-            vector=[1.0, 0.0, 0.0],
-            diversity_selection=Diversity.mmr(limit=mmr_limit, balance=0.5),
-        ),
+        vector=HybridVector.near_vector(vector=[1.0, 0.0, 0.0]),
+        diversity_selection=Diversity.mmr(limit=mmr_limit, balance=0.5),
     ).objects
     assert len(result) == mmr_limit

weaviate/collections/classes/grpc.py

@@ -617,9 +617,13 @@ def blend(
 class MMR:
     """Define MMR (Maximal Marginal Relevance) diversity selection.
 
+    Not supported for multi-vector indexes.
+
     Args:
-        limit: Optional number of candidates to consider for diversification.
-        balance: Optional MMR lambda in [0.0, 1.0] — 1.0 is pure relevance, 0.0 is pure diversity.
+        limit: Number of objects to select. The server requires it: it must be at least 1 and
+            no larger than the query's own `limit` when one is set.
+        balance: Optional MMR lambda in [0.0, 1.0]. 1.0 ranks purely by relevance to the query,
+            0.0 purely by dissimilarity to the objects already selected.
     """
 
     limit: Optional[int] = None
@@ -636,9 +640,13 @@ def __init__(self) -> None:
     def mmr(limit: Optional[int] = None, balance: Optional[float] = None) -> MMR:
         """Maximal Marginal Relevance diversity selection.
 
+        Not supported for multi-vector indexes.
+
         Args:
-            limit: Number of candidates to consider for diversification.
-            balance: MMR lambda in [0.0, 1.0] — 1.0 pure relevance, 0.0 pure diversity.
+            limit: Number of objects to select. The server requires it: it must be at least 1
+                and no larger than the query's own `limit` when one is set.
+            balance: MMR lambda in [0.0, 1.0]. 1.0 ranks purely by relevance to the query,
+                0.0 purely by dissimilarity to the objects already selected.
         """
         return MMR(limit=limit, balance=balance)
 

weaviate/collections/queries/hybrid/generate/executor.py

@@ -491,7 +491,7 @@ def hybrid(
             group_by: How the results should be grouped by a specific property.
             rerank: How the results should be reranked. NOTE: A `rerank-*` module must be enabled for this functionality to work.
             boost: A `Boost` that re-scores the search candidates to promote or demote objects without removing them.
-            diversity_selection: An `MMR` selection applied to the fused results to diversify the returned objects.
+            diversity_selection: An `MMR` selection applied to the fused results to diversify the returned objects. Requires Weaviate >= 1.38.6.
             target_vector: The name of the vector space to search in for named vector configurations. Required if multiple spaces are configured.
             include_vector: Whether to include the vector in the results. If not specified, this is set to False.
             return_metadata: The metadata to return for each object, defaults to `None`.

weaviate/collections/queries/hybrid/query/executor.py

@@ -427,7 +427,7 @@ def hybrid(
             group_by: How the results should be grouped by a specific property.
             rerank: How the results should be reranked. NOTE: A `rerank-*` module must be enabled for this functionality to work.
             boost: A `Boost` that re-scores the search candidates to promote or demote objects without removing them.
-            diversity_selection: An `MMR` selection applied to the fused results to diversify the returned objects.
+            diversity_selection: An `MMR` selection applied to the fused results to diversify the returned objects. Requires Weaviate >= 1.38.6.
             target_vector: The name of the vector space to search in for named vector configurations. Required if multiple spaces are configured.
             include_vector: Whether to include the vector in the results. If not specified, this is set to False.
             return_metadata: The metadata to return for each object, defaults to `None`.