baseline #4
Comments
Has your problem been solved? I encountered the same issue. The code seems incomplete and it appears that some files are missing. |
|
Thank you for your email and for your interest in our work.
Regarding your request for the experimental code, I’m pleased to inform you that we have completed fixing the code, and it is now fully functional and reproducible. If you encounter any issues while trying to replicate our results, please don’t hesitate to reach out to us. We’ll be more than happy to assist you.
As for the baseline-related code you mentioned, I would like to direct your attention to another work from our lab, LLaRA. Our current work is built upon the foundation laid by LLaRA. We followed the same settings and used the same dataset as described in LLaRA. You might find the baseline-related code and additional insights in the LLaRA repository, which could be beneficial for your research.
Please let me know if you need any further information or assistance.
Best regards
JOKER
***@***.***
…------------------ 原始邮件 ------------------
发件人: "AkaliKong/iLoRA" ***@***.***>;
发送时间: 2025年2月1日(星期六) 上午10:05
***@***.***>;
***@***.***>;
主题: Re: [AkaliKong/iLoRA] baseline (Issue #4)
Dear Author,
In order to better understand your approach and implement it in my research, I sincerely request if you could share the relevant experimental code or its implementation. I believe that these codes will greatly assist my research work and allow me to delve deeper into learning and exploring the related technologies. For example, how the dataset is processed, as well as the baseline-related code.
Has your problem been solved? I encountered the same issue. The code seems incomplete and it appears that some files are missing.
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>
|
Thank you very much for the author's prompt response. The code can now be run. |
|
iLoRA的作者，您好：
感谢你昨天的积极回复，昨天我成功运行了代码，但是valid的metric一直是0，非常奇怪，其实同样的问题也出现在了LLaRA的实验复现上，根据LLaRA提供的checkpoint，我inference后能够得到与其一致的结果，但是当我使用train命令的时候反而复现不出相同的结果，iLoRA也遇到了相同的问题，而且valid的metric矩阵输出我发现他是没有generate任何结果，这一点我非常疑惑，如果它输出错了，还可以理解没有结果属实有点难以理解。
在我的实验过程中，由于显卡有其他进程再跑，因此我保持了accumulate grad batch不变为16，但是将batchsize从8改为了4，想请问一下，是否batch size会影响实验运行结果，valid metric没有输出结果的原因可能是什么，麻烦作者有空能够帮我解答一下疑惑，实在是万分感谢了。
Dear author of iLoRA,
Thank you for your prompt response yesterday. I managed to run the code successfully, but the valid metric has always been 0, which is quite strange. Actually, the same issue occurred when I was reproducing the experiments of LLaRA. According to the checkpoint provided by LLaRA, I could obtain consistent results after inference, but when I used the train command, I couldn't reproduce the same results. The same problem also happened with iLoRA. Moreover, I found that the valid metric matrix output didn't generate any results, which is very confusing. If the output was wrong, it could be understandable, but there being no results at all is quite perplexing.
During my experiments, since there were other processes running on the GPU, I kept the accumulate grad batch unchanged at 16, but changed the batch size from 8 to 4. I would like to ask if the batch size could affect the experimental results, and what might be the reason for the valid metric not outputting any results. I would be extremely grateful if you could help me solve this confusion when you have time. Thank you very much.
…------------------ 原始邮件 ------------------
发件人: "AkaliKong/iLoRA" ***@***.***>;
发送时间: 2025年2月1日(星期六) 中午12:02
***@***.***>;
***@***.******@***.***>;
主题: Re: [AkaliKong/iLoRA] baseline (Issue #4)
Thank you for your email and for your interest in our work. 
Regarding your request for the experimental code, I’m pleased to inform you that we have completed fixing the code, and it is now fully functional and reproducible. If you encounter any issues while trying to replicate our results, please don’t hesitate to reach out to us. We’ll be more than happy to assist you.
As for the baseline-related code you mentioned, I would like to direct your attention to another work from our lab, LLaRA. Our current work is built upon the foundation laid by LLaRA. We followed the same settings and used the same dataset as described in LLaRA. You might find the baseline-related code and additional insights in the LLaRA repository, which could be beneficial for your research.
Please let me know if you need any further information or assistance.
Best regards
JOKER
***@***.***
 
------------------ 原始邮件 ------------------
发件人: "AkaliKong/iLoRA" ***@***.***>;
发送时间: 2025年2月1日(星期六) 上午10:05
***@***.***>;
***@***.***>;
主题: Re: [AkaliKong/iLoRA] baseline (Issue #4)
Dear Author,
In order to better understand your approach and implement it in my research, I sincerely request if you could share the relevant experimental code or its implementation. I believe that these codes will greatly assist my research work and allow me to delve deeper into learning and exploring the related technologies. For example, how the dataset is processed, as well as the baseline-related code.
Has your problem been solved? I encountered the same issue. The code seems incomplete and it appears that some files are missing.
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
|
尊敬的作者您好： |
|
感谢各位对我们工作的关注，我会尽快检查您所提到的问题的根源并进行更新。 |
|
Thank you for bringing this matter to my attention. I sincerely apologize if this situation has affected your research progress. Based on my initial assessment, the issue might be related to the increased complexity of the ilora structure, which could require adjustments to the learning rate. I believe using a smaller learning rate might resolve the problem you've encountered. Another potential solution would be to implement gradient clipping. I will carefully consider the issue you've raised and will update my GitHub repository as soon as possible with the necessary adjustments. Thank you for your understanding and patience. |
Dear Author,
In order to better understand your approach and implement it in my research, I sincerely request if you could share the relevant experimental code or its implementation. I believe that these codes will greatly assist my research work and allow me to delve deeper into learning and exploring the related technologies. For example, how the dataset is processed, as well as the baseline-related code.
The text was updated successfully, but these errors were encountered: