From a3e4857ab3091edb3e978034fd9679519e1fdced Mon Sep 17 00:00:00 2001 From: Fan Zhang Date: Sat, 29 Apr 2023 04:44:22 +0800 Subject: [PATCH] fix-step3-readme (#286) * fix-step3-readme * Update README.md --------- Co-authored-by: Zhewei Yao --- .../step3_rlhf_finetuning/training_scripts/README.md | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-) diff --git a/applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/training_scripts/README.md b/applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/training_scripts/README.md index bc5143946..e1b2b306f 100644 --- a/applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/training_scripts/README.md +++ b/applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/training_scripts/README.md @@ -7,9 +7,6 @@ If you don't have step 1 and step 2 models. You may simply try ``` bash --actor_model_name_or_path facebook/opt-1.3b --critic_model_name_or_path facebook/opt-350m ``` -⚡⚡⚡ When you use above script, please make sure you comment out the following such that it won't load the model weight from previous paths. -```bash -applications/DeepSpeed-Chat/training/utils/model/model_utils.py#L60 -``` +⚡⚡⚡ When you use above script, please make sure you modify parameter `rlhf_training` to False when calling the `create_critic_model` function twice in [rlhf_engine.py](./../../step3_rlhf_finetuning/rlhf_engine.py) such that it won't load the model weight from previous paths. For the models we support, please see [our landing page](./../../../README.md#-supported-models-)