About the scripts execute order

[Veluriyam]

bash scripts/train_llama.sh [2024-09-27 16:58:30,082] [estimate.py:67] pretrained_model_path: /root/yp/LLM-Safeguard-main/code/meta-llama/Llama-2-7b-chat-hf [2024-09-27 16:58:30,082] [estimate.py:67] config: sampling [2024-09-27 16:58:30,082] [estimate.py:67] output_path: ./estimations [2024-09-27 16:58:30,082] [estimate.py:67] system_prompt_type: all [2024-09-27 16:58:30,082] [estimate.py:67] n_splits: 10 [2024-09-27 16:58:30,089] [utils.py:94] GPU 0: 0.27 GB / 32.00 GB [2024-09-27 16:58:30,095] [utils.py:94] GPU 1: 15.65 GB / 32.00 GB [2024-09-27 16:58:30,224] [estimate.py:89] /root/yp/LLM-Safeguard-main/code/meta-llama/Llama-2-7b-chat-hf [2024-09-27 16:58:30,226] [estimate.py:98] Running harmful [2024-09-27 16:58:30,694] [estimate.py:124] Running harmless [2024-09-27 16:58:30,922] [utils.py:46] File eval_results/sampling/Llama-2-7b-chat-hf_custom.csv does not exist, exiting [2024-09-27 16:58:34,948] [train.py:107] pretrained_model_path: /root/yp/LLM-Safeguard-main/code/meta-llama/Llama-2-7b-chat-hf [2024-09-27 16:58:34,948] [train.py:107] config: sampling [2024-09-27 16:58:34,948] [train.py:107] system_prompt_type: all [2024-09-27 16:58:34,948] [train.py:107] prompt_length: default [2024-09-27 16:58:34,948] [train.py:107] output_path: ./trained_prompts [2024-09-27 16:58:34,948] [train.py:107] ablate_norm: False [2024-09-27 16:58:34,948] [train.py:107] ablate_refu: False [2024-09-27 16:58:34,948] [train.py:107] ablate_harm: False Hello, fllow your Readme, after I bash scripts/forward.sh with forward_harmless.sh, when I bash scripts/train_llama.sh it makes the error that can't find eval_results/sampling/Llama-2-7b-chat-hf_custom.csv, and I noted that evaluate.py generates the . /eval_results file in order to get the scores in utils.py . So is there something wrong with my execution order, do I need to execute forward.sh and forward_less.sh first, then execute evaluate.py to get the scores so that I can execute estimate.py to generate the csv file, and if so why is there no such step in train_meta.sh? 你好，看了你的 Readme，在我执行 scripts/forward.sh和forward_harmless.sh 之后，当我执行 scripts/train_llama.sh 时，报错 eval_results/sampling/Llama-2-7b-chat-hf_custom.csv ，我注意到 evaluate.py 会生成 ./eval_results 文件，以便在 utils.py 中获取分数。那么我的执行顺序是否有问题，我是否需要先执行 forward.sh 和 forward_less.sh，然后执行 evaluate.py 获得分数，这样才能执行 estimate.py 生成 csv 文件，如果是这样，为什么 train_meta.sh 中没有这样的步骤？