fix dim() (embedding dimensions) method for qdrant cloud dense encoders

vector_search/encoders/qdrant_cloud.py

@@ -1,5 +1,6 @@
 import logging
 
+import litellm
 import tiktoken
 from django.conf import settings
 from qdrant_client import models
@@ -19,7 +20,9 @@ class QdrantCloudEncoder(BaseEncoder):
     def __init__(self, model_name):
         self.model_name = model_name
         try:
-            self.token_encoding_name = tiktoken.encoding_name_for_model(model_name)
+            self.token_encoding_name = tiktoken.encoding_name_for_model(
+                self.model_short_name()
+            )
         except KeyError:
             msg = f"Model {model_name} not found in tiktoken. defaulting to None"
             log.warning(msg)
@@ -42,3 +45,10 @@ def get_embedding(self, texts):
             )
             for text in texts
         ]
+
+    def dim(self):
+        """
+        Return the dimension of the embeddings
+        """
+        info = litellm.get_model_info(self.model_short_name())
+        return info["output_vector_size"]

vector_search/utils.py

@@ -900,6 +900,7 @@ def vector_search(  # noqa: PLR0913
     encoder_sparse = sparse_encoder()
 
     search_filter = qdrant_query_conditions(params, collection_name=search_collection)
+
     prefetch_multiplier = settings.VECTOR_HYBRID_SEARCH_PREFETCH_MULTIPLIER
     prefetch_max_limit = settings.VECTOR_HYBRID_SEARCH_PREFETCH_MAX_LIMIT
     prefetch_limit = min((offset + limit) * prefetch_multiplier, prefetch_max_limit)