From 5066f67c8e177319113dabd9ed97455449388e17 Mon Sep 17 00:00:00 2001 From: randaller Date: Sat, 11 Mar 2023 18:36:00 +0300 Subject: [PATCH] Update README.md --- README.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index affc359..ad9cd11 100644 --- a/README.md +++ b/README.md @@ -102,8 +102,8 @@ For sure, you may play with all the values to get different outputs. One may modify these hyperparameters straight in the code. But it is better to leave the defaults in code and set the parameters of experiments in the launch line. ``` -# Run with top_p sampler, with temperature 0.75, with top_p value 0.9, repetition penalty disabled -python example-chat.py ./model ./tokenizer/tokenizer.model 0.75 0.9 0 1.0 top_p +# Run with top_p sampler, with temperature 0.75, with top_p value 0.95, repetition penalty disabled +python example-chat.py ./model ./tokenizer/tokenizer.model 0.75 0.95 0 1.0 top_p # Run with top_k sampler, with temperature 0.7, with top_k value 40, default repetition penalty value python example-chat.py ./model ./tokenizer/tokenizer.model 0.7 0.0 40 1.17 top_k