Embeddings API scores (#671)

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

trustgraph-flow/trustgraph/embeddings/document_embeddings/embeddings.py

@@ -66,13 +66,13 @@ class Processor(FlowProcessor):
                 )
             )
 
-            # vectors[0] is the vector set for the first (only) text
-            vectors = resp.vectors[0] if resp.vectors else []
+            # vectors[0] is the vector for the first (only) text
+            vector = resp.vectors[0] if resp.vectors else []
 
             embeds = [
                 ChunkEmbeddings(
                     chunk_id=v.document_id,
-                    vectors=vectors,
+                    vector=vector,
                 )
             ]
 

trustgraph-flow/trustgraph/embeddings/fastembed/processor.py

@@ -59,11 +59,8 @@ class Processor(EmbeddingsService):
         # FastEmbed processes the full batch efficiently
         vecs = list(self.embeddings.embed(texts))
 
-        # Return list of vector sets, one per input text
-        return [
-            [v.tolist()]
-            for v in vecs
-        ]
+        # Return list of vectors, one per input text
+        return [v.tolist() for v in vecs]
 
     @staticmethod
     def add_args(parser):

trustgraph-flow/trustgraph/embeddings/graph_embeddings/embeddings.py

@@ -72,10 +72,10 @@ class Processor(FlowProcessor):
             entities = [
                 EntityEmbeddings(
                     entity=entity.entity,
-                    vectors=vectors,  # Vector set for this entity
+                    vector=vector,
                     chunk_id=entity.chunk_id,  # Provenance: source chunk
                 )
-                for entity, vectors in zip(v.entities, all_vectors)
+                for entity, vector in zip(v.entities, all_vectors)
             ]
 
             # Send in batches to avoid oversized messages

trustgraph-flow/trustgraph/embeddings/ollama/processor.py

@@ -43,11 +43,8 @@ class Processor(EmbeddingsService):
             input = texts
         )
 
-        # Return list of vector sets, one per input text
-        return [
-            [embedding]
-            for embedding in embeds.embeddings
-        ]
+        # Return list of vectors, one per input text
+        return list(embeds.embeddings)
 
     @staticmethod
     def add_args(parser):

trustgraph-flow/trustgraph/embeddings/row_embeddings/embeddings.py

@@ -208,7 +208,7 @@ class Processor(CollectionConfigHandler, FlowProcessor):
             all_vectors = await flow("embeddings-request").embed(texts=texts)
 
             # Pair results with metadata
-            for text, (index_name, index_value), vectors in zip(
+            for text, (index_name, index_value), vector in zip(
                 texts, metadata, all_vectors
             ):
                 embeddings_list.append(
@@ -216,7 +216,7 @@ class Processor(CollectionConfigHandler, FlowProcessor):
                         index_name=index_name,
                         index_value=index_value,
                         text=text,
-                        vectors=vectors  # Vector set for this text
+                        vector=vector
                     )
                 )