Training Data Availability and Scaling Coefficient Hyperparameters

[JinXins]

Hi, @Littleor

Great work on ExpertMerging! I’m also very interested in this topic and planning to conduct related research.
I have a couple of quick questions:

• Could you share the merging scaling coefficients used for other baseline methods in your paper? I didn’t find them mentioned in the paper.

• Do you plan to release your training data soon? It would be very helpful for others to reproduce and implement your work.

Thanks a lot for your time and contribution!

Best regards,
Xin

[Littleor]

Thank you for your interest in ExpertMerging!

1. Scaling coefficients for baselines: We followed the method in https://arxiv.org/pdf/2505.19892, likewise searching for the optimal merging coefficient λ within [0.1, 0.3, 0.5, 0.7, 1.0, 1.5].
2. Training data: We used the dataset from https://github.com/WalkerWorldPeace/MLLMerging/, sampled the specified number of examples as described in paper, and then performed the merge; you can obtain this data directly from the repository.