From e4f6e6b6fb97372c2a58fce3a59d5241caf39518 Mon Sep 17 00:00:00 2001 From: PhilipMay Date: Sun, 14 Apr 2024 07:46:57 +0200 Subject: [PATCH] add img description --- source/blog/2024/importance-of-chat-templates.md | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/source/blog/2024/importance-of-chat-templates.md b/source/blog/2024/importance-of-chat-templates.md index 7f4f30f..90c1089 100644 --- a/source/blog/2024/importance-of-chat-templates.md +++ b/source/blog/2024/importance-of-chat-templates.md @@ -95,6 +95,10 @@ This is exactly what [Daniel Furman](https://github.com/daniel-furman) did on th :::{figure} /\_static/img/blog-2024/model-eval-template.png :width: 550px + +Qualitative comparison of different LLMs with and without the template. +Table by [Daniel Furman](https://github.com/daniel-furman) from this +[GitHub issue conversation](https://github.com/EleutherAI/lm-evaluation-harness/issues/1098#issuecomment-1953068243). ::: The table clearly shows that the use of the (correct) template has a significant impact on the quality of the generated answers.