Help with Submitting Our Model for the Multi-Author Writing Style Change Detection Task

PAN
1

Dear @maik_froebe

I am writing to ask for your guidance on how to submit our approach for the Multi-Author Writing Style Change Detection task.

We would like to submit our code (including predict.py) along with our trained model files (e.g., .pth). Initially, we considered using the code submission approach via GitHub. However, we are aware that large files (such as .pt or .bin) are not recommended to be included directly in a Git repository.

Additionally, when verifying our TIRA installation using the tira-cli verify-installation command, we encountered some errors similar to those mentioned in a previously posted question on the forum.

docker1
docker1634×141 4.62 KB

Could you please clarify the recommended way to submit our model using the code submission approach? Specifically, we would appreciate your advice on how to handle large model files like .pt or .bin. We are concerned that using the Docker submission approach might result in an image that is too large, due to the size of our model parameters.

Thank you in advance for your help!

Best regards,
Mehrani

2

Dear Mehrani,

Thanks for reaching out and for participating in the task!

The preferred way for me would still be a code submission. If you want to make this from your local machine, you need to install docker, as the code will be embedded in the docker image. If you do not want to install docker on your machine, you can use the github action (the stepper guides you through the process). If you use Github actions, there needs to be a way to get the model into the cloud, in that case, I think it would be the preferred way to load it either from Hugging Face or some other source. If you have problems, I can help you if you invite me to the repository (my account on Github is mam10eks).

Best regards,

Maik

3

Dear Maik,
Thank you so much for your help.
Regarding that case, I’ll message you in private and ask you my questions there.
Regards,
Mehrani

1 Like