From 08895fdfff500985c8f990bc69df67982d040baa Mon Sep 17 00:00:00 2001 From: Ella Charlaix <80481427+echarlaix@users.noreply.github.com> Date: Fri, 15 Mar 2024 23:05:24 +0100 Subject: [PATCH] Fix table (#1906) --- intel-fast-embedding.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/intel-fast-embedding.md b/intel-fast-embedding.md index 693f9b8b8d..f8efd58cb5 100644 --- a/intel-fast-embedding.md +++ b/intel-fast-embedding.md @@ -149,10 +149,10 @@ Quantizing the models' weights to a lower precision introduces accuracy loss, as The table below shows the average accuracy (on multiple datasets) of each task type (MAP for Reranking, NDCG@10 for Retrieval), where `int8` is our quantized model and `fp32` is the original model (results taken from the official MTEB leaderboard). The quantized models show less than 1% error rate compared to the original model in the Reranking task and less than 1.55% in the Retrieval task. <table> -<tr><th> Model </th><th> Reranking </th><th> Retrieval </th></tr> +<tr><th> </th><th> Reranking </th><th> Retrieval </th></tr> <tr><td> -| precision | +| | | --------- | | BGE-small | | BGE-base |