Can you share the Complete Hyperparameters in different lambda?

[zhongmz]

Before asking:
try different Hyperparameters (lr , warmup-updates ...)

In the reproduction of Dual-Path, the performance of the model in the test set is good when lambda=0.01/0.05/0.1/0.2 in the en-vi dataset, However, for lambda=0.3/0.4, the bleu values in the test set are only 6.51 and 0.51, and the convergence loss of these two lambdas in the valid set is much higher than the others.

so I would like to ask you about the Complete Hyperparameters settings for different lambdas in different data sets (en-de, en-vi)?

Sorry for any interruptions.