Filter multilabel confidence keys for llm labels. Currently, the multilabel confidence may be computed for tokens that were filtered out when parsing the llm response. We don't want to include confidence for labels that were filtered out during parsing logic.
Pull Review Summary
Description
Filter multilabel confidence keys for llm labels. Currently, the multilabel confidence may be computed for tokens that were filtered out when parsing the llm response. We don't want to include confidence for labels that were filtered out during parsing logic.
Type of change
Tests
Tested confidence key filtering logic