Training the nlu

Rasa Open Source
1

I have around 3000 training sentences with 7 different entities and 4 intents

I am training the rasa only for nlu part

I trained it for 100 epocs but getting very high loss(3.86) you can see below

Epochs: 100%|██████████████████████████████████████████████████████████| 100/100 [12:56<00:00,  7.77s/it, t_loss=3.86, m_acc=0.917, i_acc=1, e_f1=0.97]

Can anyone help me to understand the mistake and decrease the loss?

If I increase the epoch it will start overfitting the data ,is there any way to check that ?

2

You have a really good accuracy though, why care about the loss?

And it is no that big to be honest.

To see what’s the best number of epochs and compare multiple pipelines, I suggest using TensorBoard.