[ENH] When embedding functions have defined default_space, use them if the user hasn't specified

chromadb/api/client.py

@@ -153,6 +153,9 @@ def create_collection(
             configuration = {}
             if embedding_function is not None:
                 configuration["embedding_function"] = embedding_function
+        else:
+            if configuration.get("embedding_function") is None:
+                configuration["embedding_function"] = embedding_function
         model = self._server.create_collection(
             name=name,
             metadata=metadata,

chromadb/api/collection_configuration.py

@@ -377,6 +377,24 @@ def create_collection_configuration_to_json(
                 "config": ef.get_config(),
             }
             register_embedding_function(type(ef))  # type: ignore
+            if hnsw_config is not None and hnsw_config.get("space") is None:
+                try:
+                    hnsw_config["space"] = ef.default_space()
+                except Exception:
+                    warnings.warn(
+                        f"default_space not supported for {ef.name()}",
+                        DeprecationWarning,
+                        stacklevel=2,
+                    )
+            if spann_config is not None and spann_config.get("space") is None:
+                try:
+                    spann_config["space"] = ef.default_space()
+                except Exception:
+                    warnings.warn(
+                        f"default_space not supported for {ef.name()}",
+                        DeprecationWarning,
+                        stacklevel=2,
+                    )
     except Exception as e:
         warnings.warn(
             f"legacy embedding function config: {e}",

chromadb/test/configurations/test_collection_configuration.py

@@ -776,3 +776,95 @@ def test_default_collection_creation(client: ClientAPI) -> None:
     ef = config.get("embedding_function")
     assert ef is not None
     assert ef.name() == "default"
+
+
+def test_default_space_hnsw(client: ClientAPI) -> None:
+    """Test that the default space is used for HNSW if not specified in config."""
+    client.reset()
+    hnsw_config_no_space: CreateHNSWConfiguration = {"ef_construction": 123}
+    coll = client.create_collection(
+        name="test_default_space_hnsw",
+        configuration={
+            "hnsw": hnsw_config_no_space,
+            "embedding_function": CustomEmbeddingFunction(),
+        },
+    )
+    config = load_collection_configuration_from_json(coll.configuration_json)
+    assert config is not None
+    hnsw_config = config.get("hnsw")
+    assert hnsw_config is not None
+    assert hnsw_config.get("space") == "cosine"  # Should match EF default
+    assert hnsw_config.get("ef_construction") == 123
+    assert config.get("spann") is None
+    ef = config.get("embedding_function")
+    assert ef is not None
+    assert ef.name() == "custom_ef"
+
+
+@pytest.mark.skipif(is_spann_disabled_mode, reason=skip_reason_spann_disabled)
+def test_default_space_spann(client: ClientAPI) -> None:
+    """Test that the default space is used for SPANN if not specified in config."""
+    client.reset()
+    spann_config_no_space: CreateSpannConfiguration = {"ef_construction": 123}
+    coll = client.create_collection(
+        name="test_default_space_spann",
+        configuration={
+            "spann": spann_config_no_space,
+            "embedding_function": CustomEmbeddingFunction(),
+        },
+    )
+    config = load_collection_configuration_from_json(coll.configuration_json)
+    assert config is not None
+    spann_config = config.get("spann")
+    assert spann_config is not None
+    assert spann_config.get("space") == "cosine"  # Should match EF default
+    assert spann_config.get("ef_construction") == 123
+    assert config.get("hnsw") is None
+    ef = config.get("embedding_function")
+    assert ef is not None
+    assert ef.name() == "custom_ef"
+
+
+def test_override_default_space_hnsw(client: ClientAPI) -> None:
+    """Test that specifying space in HNSW config overrides the EF default."""
+    client.reset()
+    hnsw_config_override: CreateHNSWConfiguration = {"space": "l2"}
+    coll = client.create_collection(
+        name="test_override_space_hnsw",
+        configuration={
+            "hnsw": hnsw_config_override,
+            "embedding_function": CustomEmbeddingFunction(),
+        },
+    )
+    config = load_collection_configuration_from_json(coll.configuration_json)
+    assert config is not None
+    hnsw_config = config.get("hnsw")
+    assert hnsw_config is not None
+    assert hnsw_config.get("space") == "l2"  # Should be overridden value
+    assert config.get("spann") is None
+    ef = config.get("embedding_function")
+    assert ef is not None
+    assert ef.name() == "custom_ef"
+
+
+@pytest.mark.skipif(is_spann_disabled_mode, reason=skip_reason_spann_disabled)
+def test_override_default_space_spann(client: ClientAPI) -> None:
+    """Test that specifying space in SPANN config overrides the EF default."""
+    client.reset()
+    spann_config_override: CreateSpannConfiguration = {"space": "l2"}
+    coll = client.create_collection(
+        name="test_override_space_spann",
+        configuration={
+            "spann": spann_config_override,
+            "embedding_function": CustomEmbeddingFunction(),
+        },
+    )
+    config = load_collection_configuration_from_json(coll.configuration_json)
+    assert config is not None
+    spann_config = config.get("spann")
+    assert spann_config is not None
+    assert spann_config.get("space") == "l2"  # Should be overridden value
+    assert config.get("hnsw") is None
+    ef = config.get("embedding_function")
+    assert ef is not None
+    assert ef.name() == "custom_ef"

chromadb/types.py

@@ -16,8 +16,6 @@
 from chromadb.serde import BaseModelJSONSerializable
 from chromadb.api.collection_configuration import (
     CollectionConfiguration,
-    HNSWConfiguration,
-    SpannConfiguration,
     collection_configuration_to_json,
     load_collection_configuration_from_json,
 )
@@ -155,8 +153,8 @@ def get_configuration(self) -> CollectionConfiguration:
                 stacklevel=2,
             )
             return CollectionConfiguration(
-                hnsw=HNSWConfiguration(),
-                spann=SpannConfiguration(),
+                hnsw=None,
+                spann=None,
                 embedding_function=None,
             )
 
@@ -175,11 +173,11 @@ def get_model_fields(self) -> Dict[Any, Any]:
     @override
     def from_json(cls, json_map: Dict[str, Any]) -> Self:
         """Deserializes a Collection object from JSON"""
-        configuration: CollectionConfiguration = {
-            "hnsw": {},
-            "spann": {},
-            "embedding_function": None,
-        }
+        configuration = CollectionConfiguration(
+            hnsw=None,
+            spann=None,
+            embedding_function=None,
+        )
         try:
             configuration_json = json_map.get("configuration_json", None)
             configuration = load_collection_configuration_from_json(configuration_json)