From 46081c055dc92eb458b6efaa216d7ea3eeb29fb0 Mon Sep 17 00:00:00 2001 From: Nirant Kasliwal Date: Mon, 4 Mar 2024 12:24:25 +0530 Subject: [PATCH] Remove the tag --- qdrant-landing/content/articles/binary-quantization-openai.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/qdrant-landing/content/articles/binary-quantization-openai.md b/qdrant-landing/content/articles/binary-quantization-openai.md index 5478e55ff..e21523006 100644 --- a/qdrant-landing/content/articles/binary-quantization-openai.md +++ b/qdrant-landing/content/articles/binary-quantization-openai.md @@ -102,7 +102,7 @@ Here are some key observations, which analyzes the impact of rescoring (`True` o - For the `text-embedding-3-large` model with 3072 dimensions, rescoring boosts the accuracy from an average of about 76-77% without rescoring to 97-99% with rescoring, depending on the search limit and oversampling rate. - The accuracy improvement with increased oversampling is more pronounced when rescoring is enabled, indicating a better utilization of the additional binary codes in refining search results. - With the `text-embedding-3-small` model at 512 dimensions, accuracy increases from around 53-55% without rescoring to 71-91% with rescoring, highlighting the significant impact of rescoring, especially at lower dimensions. - - For higher dimension models (such as text-embedding-3-large with 3072 dimensions), + In contrast, for lower dimension models (such as text-embedding-3-small with 512 dimensions), the incremental accuracy gains from increased oversampling levels are less significant, even with rescoring enabled. This suggests a diminishing return on accuracy improvement with higher oversampling in lower dimension spaces. 3. **Influence of Search Limit**: