So weird training result

[alexnotes]

When i trained the model with these parameters, the result is fine. Normally, F1 Score is up to 0.81 or even better.

glove.6B.100d.txt
feature_size=100
hidden_size=200
filter_size=800

But when i tried the following hyper-parameter, F1 score is always less than 0.1. Specifically, it is 0.042365085 and the model only predicted B-V label. No any other labels. It is so weird. What's the problem?

glove.6B.200d.txt
feature_size=200
hidden_size=400
filter_size=1600

[xiaocainiao]

similar to the problem I encountered