JIAYI LIAO

The GPU-util mentioned above is when I run the examples/02_model_content_based_filtering/dkn_deep_dive.ipynb.

Yes. ------------------ 原始邮件 ------------------ 发件人: "microsoft/recommenders" ***@***.***>; 发送时间: 2021年9月16日(星期四) 晚上9:28 ***@***.***>; ***@***.******@***.***>; 主题: Re: [microsoft/recommenders] [ASK] How to run with docker in GPU environment (#1526) do you see any gpu available by running this...

Thanks for the code. I'm also wondering about the same question. Are the default hyper-parameters that you used to report the results in your paper?

Thanks for your attention to our work. As you mentioned, batch size and precision will affect the stability of training.

Thanks for your attention to our work. Perhaps the problem is that the Llama model itself takes up a certain amount of GPU memory.

感谢您对我们工作的关注！ main.py文件里开头import了SASRec等传统推荐模型，您可以检查一下python运行环境中包导入的路径配置。

Hi, thanks for your attention to our work. You could try to ensure that the packages mentioned in the requirements.txt file are correctly installed on the A100 machine (especially the...

Thanks for your attention to our work! We have released the datasets and checkpoints.

Hi~ Thanks for your attention to our work. The termination of the process doesn't seem to be due to a lack of checkpoints, as the training process doesn't require LLaRA...

It seems your program was killed while executing the line of code ```self.llama_model = LlamaForCausalLM.from_pretrained(llm_path, torch_dtype=torch.bfloat16)``` in mode/model_interface.py. Perhaps you could try directly loading the LLaMA-2 7B Hugging Face version...