From 219e6c2ac9c230d62500528721c4660b1b474e6c Mon Sep 17 00:00:00 2001 From: xiamengzhou Date: Mon, 15 Jul 2024 09:26:20 -0400 Subject: [PATCH] Update README.md --- README.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/README.md b/README.md index c3ae28e..59e14ed 100644 --- a/README.md +++ b/README.md @@ -57,6 +57,8 @@ Our released Llama3 models use the initial version of the Llama3 tokenizer (prio *Notably, if you are training Llama3 and evaluating the trained models on AlpacaEval 2 and Arena-Hard using the templates provided in this repo, please make sure to use the pre-update Llama3 tokenizer (i.e., the one before the PR).* +### Reproducing AlpacaEval 2 numbers +Please make sure that you use `AlpacaEval==0.6.2` for successfully reproducing AlpacaEval 2 results. ### Adding an extra SFT loss The [CPO_SIMPO](https://github.com/fe1ixxu/CPO_SIMPO/tree/main) repository did preliminary experiments and observed that in some cases, adding an additional SFT loss can help improve results. In our own experiments, the SFT regularization helps preserve the reasoning ability (e.g., GSM8K) but degrades chat performance. If you'd like to apply SFT regularization, you can set `sft_weight` to be a positive value (by default it's 0).