overfitting when training on cifar10

[lowestbuaaer]

I trained a model on cifar10 with below settings. MODEL_FLAGS="--image_size 32 --num_channels 128 --num_res_blocks 3 --learn_sigma True --dropout 0.1" DIFFUSION_FLAGS="--diffusion_steps 4000 --noise_schedule linear" TRAIN_FLAGS="--lr 1e-4 --batch_size 16" but it overfits at 500k iterations. optimal point locates at about 300k iterations.(sample 100 steps, FID 4.7) Does anyone know possible reasons?

[99-WSJ]

maybe you check the dropout== 0.3 in readme,

[yug125lk]

Hi, Did you solve it?