Embeddings API scores (#671)

- Put scores in all responses
- Remove unused 'middle' vector layer. Vector of texts -> vector of (vector embedding)

trustgraph-base/trustgraph/schema/knowledge/embeddings.py

@@ -11,7 +11,7 @@ from ..core.topic import topic
 @dataclass
 class EntityEmbeddings:
     entity: Term | None = None
-    vectors: list[list[float]] = field(default_factory=list)
+    vector: list[float] = field(default_factory=list)
     # Provenance: which chunk this embedding was derived from
     chunk_id: str = ""
 
@@ -28,7 +28,7 @@ class GraphEmbeddings:
 @dataclass
 class ChunkEmbeddings:
     chunk_id: str = ""
-    vectors: list[list[float]] = field(default_factory=list)
+    vector: list[float] = field(default_factory=list)
 
 # This is a 'batching' mechanism for the above data
 @dataclass
@@ -44,7 +44,7 @@ class DocumentEmbeddings:
 @dataclass
 class ObjectEmbeddings:
     metadata: Metadata | None = None
-    vectors: list[list[float]] = field(default_factory=list)
+    vector: list[float] = field(default_factory=list)
     name: str = ""
     key_name: str = ""
     id: str = ""
@@ -56,7 +56,7 @@ class ObjectEmbeddings:
 @dataclass
 class StructuredObjectEmbedding:
     metadata: Metadata | None = None
-    vectors: list[list[float]] = field(default_factory=list)
+    vector: list[float] = field(default_factory=list)
     schema_name: str = ""
     object_id: str = ""  # Primary key value
     field_embeddings: dict[str, list[float]] = field(default_factory=dict)  # Per-field embeddings
@@ -72,7 +72,7 @@ class RowIndexEmbedding:
     index_name: str = ""              # The indexed field name(s)
     index_value: list[str] = field(default_factory=list)  # The field value(s)
     text: str = ""                    # Text that was embedded
-    vectors: list[list[float]] = field(default_factory=list)
+    vector: list[float] = field(default_factory=list)
 
 @dataclass
 class RowEmbeddings:

trustgraph-base/trustgraph/schema/services/llm.py

@@ -34,7 +34,7 @@ class EmbeddingsRequest:
 @dataclass
 class EmbeddingsResponse:
     error: Error | None = None
-    vectors: list[list[list[float]]] = field(default_factory=list)
+    vectors: list[list[float]] = field(default_factory=list)
 
 ############################################################################
 

trustgraph-base/trustgraph/schema/services/query.py

@@ -9,15 +9,21 @@ from ..core.topic import topic
 
 @dataclass
 class GraphEmbeddingsRequest:
-    vectors: list[list[float]] = field(default_factory=list)
+    vector: list[float] = field(default_factory=list)
     limit: int = 0
     user: str = ""
     collection: str = ""
 
+@dataclass
+class EntityMatch:
+    """A matching entity from a semantic search with similarity score"""
+    entity: Term | None = None
+    score: float = 0.0
+
 @dataclass
 class GraphEmbeddingsResponse:
     error: Error | None = None
-    entities: list[Term] = field(default_factory=list)
+    entities: list[EntityMatch] = field(default_factory=list)
 
 ############################################################################
 
@@ -44,15 +50,21 @@ class TriplesQueryResponse:
 
 @dataclass
 class DocumentEmbeddingsRequest:
-    vectors: list[list[float]] = field(default_factory=list)
+    vector: list[float] = field(default_factory=list)
     limit: int = 0
     user: str = ""
     collection: str = ""
 
+@dataclass
+class ChunkMatch:
+    """A matching chunk from a semantic search with similarity score"""
+    chunk_id: str = ""
+    score: float = 0.0
+
 @dataclass
 class DocumentEmbeddingsResponse:
     error: Error | None = None
-    chunk_ids: list[str] = field(default_factory=list)
+    chunks: list[ChunkMatch] = field(default_factory=list)
 
 document_embeddings_request_queue = topic(
     "document-embeddings-request", qos='q0', tenant='trustgraph', namespace='flow'
@@ -76,7 +88,7 @@ class RowIndexMatch:
 @dataclass
 class RowEmbeddingsRequest:
     """Request for row embeddings semantic search"""
-    vectors: list[list[float]] = field(default_factory=list)  # Query vectors
+    vector: list[float] = field(default_factory=list)  # Query vector
     limit: int = 10                         # Max results to return
     user: str = ""                          # User/keyspace
     collection: str = ""                    # Collection name