cr

gpt_index/indices/postprocessor/node.py

@@ -61,18 +61,16 @@ def postprocess_nodes(
  ) -> List[Node]:
  """Postprocess nodes."""
  extra_info = extra_info or {}
+ similarity_tracker = extra_info.get("similarity_tracker", None)
+ if similarity_tracker is None:
+ return nodes
+ sim_cutoff_exists = (
+ similarity_tracker is not None and self.similarity_cutoff is not None
+ )
+
  new_nodes = []
  for node in nodes:
  should_use_node = True
- similarity_tracker = extra_info.get("similarity_tracker")
- if similarity_tracker is None:
- raise ValueError(
- "Similarity tracker is required for similarity postprocessor."
- )
- sim_cutoff_exists = (
- similarity_tracker is not None and self.similarity_cutoff is not None
- )
-
  if sim_cutoff_exists:
  similarity = cast(SimilarityTracker, similarity_tracker).find(node)
  if similarity is None:

gpt_index/indices/query/knowledge_graph/query.py

@@ -176,8 +176,11 @@ def _get_nodes_for_response(
  self.index_struct.text_chunks[idx] for idx in sorted_chunk_indices
  ]
  # filter sorted nodes
+ postprocess_info = {"similarity_tracker": similarity_tracker}
  for node_processor in self.node_preprocessors:
- sorted_nodes = node_processor.postprocess_nodes(sorted_nodes)
+ sorted_nodes = node_processor.postprocess_nodes(
+ sorted_nodes, postprocess_info
+ )
 
  # TMP/TODO: also filter rel_texts as nodes until we figure out better
  # abstraction

tests/indices/knowledge_graph/test_base.py

@@ -232,8 +232,8 @@ def test_query_similarity(
  assert isinstance(response.extra_info, dict)
  assert len(response.extra_info["kg_rel_texts"]) == 2
 
- # Filters out all embeddings
+ # Filters embeddings
  try:
- response = index.query("foo", similarity_cutoff=-1.0)
+ response = index.query("foo", similarity_cutoff=10000000)
  except ValueError as e:
  assert str(e) == "kg_rel_map must be found in at least one Node."