Sharing Large Pre-Trained Models

89 views
Skip to first unread message

Lucas Erlacher

unread,
Jun 16, 2025, 10:10:27 AMJun 16
to physionet-challenges

Dear Challenge Organizers,


I have two questions:

- we are using a pre-trained model (that we are fine-tuning) which is about 300MB in size. given that GitHub limits each individual file to not exceed 100MB how would you recommend us to share these pre-trained weights s.t. you can run our training on your end?

- I have just seen that in your test dataset you will use different sources than "code_15", "ptbxl" and "samitrop". since we are currently using this source information in our training code we need to understand what exactly will change: will the format of the header itself be different, or will the format remain the same with only the source string differing?

PhysioNet Challenge

unread,
Jun 16, 2025, 10:14:09 AMJun 16
to physionet-challenges
Dear Lucas,

There are multiple options for uploading larger model weights, including using Git LFS, using GitLab, or downloading model weights while building your Docker image. Some teams may also decide not to save all of the model weights, or to save them with fewer bits.

(Again, we encourage teams not upload models to avoid submitting working training code or the provide training resource constraints. We will verify that the submitted training code actually runs and learns from the training data for rankings and prize eligibility.)

The data in the validation and test sets are from different sources than the data in the training set. The validation and test data are formatted exactly the same as the training data, except for the labels or sources; we do not include the labels or data sources in the validation and test sets so that teams can infer the labels and to better assess how teams generalize to new data sources.
https://moody-challenge.physionet.org/2025/#data

Best,
Matt
(On behalf of the Challenge team.)

Please post questions and comments in the forum. However, if your question reveals information about your entry, then please email info at physionetchallenge.org. We may post parts of our reply publicly if we feel that all Challengers should benefit from it. We will not answer emails about the Challenge to any other address. This email is maintained by a group. Please do not email us individually.
Reply all
Reply to author
Forward
0 new messages