How to train select agent?

[cingtiye]

Thank you for this great work! I'm every interested in your work.

When I reproduce select agent model training, I faced some problems. I trained a binary classifier using BCE loss. My prompt is shown in the following picture. I tried to train LLaMA3.1 and Qwen2.5 in BIRD train set, but I failed. My model quickly converged on the training set, but overfited on the validation set.

Could you please tell me how you trained this model? and thanks in advance