From 43cf32194253303ff0312470d4983acd8d5dcbe9 Mon Sep 17 00:00:00 2001 From: xintoteai <160626233+xintoteai@users.noreply.github.com> Date: Thu, 10 Apr 2025 04:17:45 -0700 Subject: [PATCH] Added similarity scores in reference chunks (#6918) - Returning 3 similarity scores to the chat completion's `reference` field. It gives the user more transparency and added flexibility to display/rerank the reference when needed Co-authored-by: Yingfeng --- rag/prompts.py | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/rag/prompts.py b/rag/prompts.py index 7433d404..489ead2f 100644 --- a/rag/prompts.py +++ b/rag/prompts.py @@ -39,7 +39,10 @@ def chunks_format(reference): "dataset_id": get_value(chunk, "kb_id", "dataset_id"), "image_id": get_value(chunk, "image_id", "img_id"), "positions": get_value(chunk, "positions", "position_int"), - "url": chunk.get("url") + "url": chunk.get("url"), + "similarity": chunk.get("similarity"), + "vector_similarity": chunk.get("vector_similarity"), + "term_similarity": chunk.get("term_similarity"), } for chunk in reference.get("chunks", [])]