Fei Gao

Hi, Thanks for your contribution! I cloned your codes ([python3 version](https://github.com/johnny12150/DySAT_python3)) and trained the model using **Enron_new** you provided. When setting learning rate=0.001 (default value), the final AUC is around...

AUC decreases A LOT after re-generating cached data

1

Hi, When I remove the cached data you provided from property prediction datasets , and generate them by myself using your codes, the AUC of property prediction decreases a lot....

各位好， 我初次尝试LLM相关的技术，在将QAnything部署到本地后，利用 `bash scripts/run_for_openai_api_in_M1_mac.sh`运行，并且修改了其中的api相关的配置，尝试利用GLM 4 的api，修改的结果如下： `bash scripts/base_run.sh -o -s "M1mac" -w 4 -m 19530 -q 8777 -b 'https://open.bigmodel.cn/api/paas/v4/' -k '*******' -n 'glm-4' -l '4096'` 但是在提问时，会返回： `Error code: 400 -...

Now I provide you the missing PREPROCESSING codes.

23

Dear all, After days of effort, [I finally completed the missing parts of this repository](https://github.com/FeiGSSS/Eadro), i.e. the preprocessing codes from raw datasets to model inputs, as requested by many colleagues:...

为什么 d_NL[(count++) * N + n1] 是合并访问的？

1

老师您好， 我不明白为什么 `d_NL[(count++) * N + n1]` 一定能保证合并内存访问？ 每个 n1 的 count 增长不是同步的，也就是说每个线程对于 d_NL 的访存地址可能差距好几个 `N`。那为什么这样能实现合并访问呢？ 还是说，只能保证 count = 0 时是合并的？