Training on large intent examples

I want to train a bot with large number of data, which may lead to out of memory or higher training time. Can you please suggest me any quick and effiecient ways to train my bot. Or can I train into chunks so to avoid those errors. please suggest me any ideas. Example: I have 10000 data per intent and there will be 5-6 intents above for my bot training. However entities will be very less Thanks

If computation power is an issue- train on chunks and finetune. Say, 100 examples per intent in the first training and then increase examples and finetune your model using -

rasa train --finetune --epoch-fraction 0.2

For more info, refer - Finetune

1 Like

What is the significance of using 10,000 data for 5-6 intents? As for training examples its recommended to use at least 10-20 training examples so 6 intents 20 training examples = 120 training examples (data) and training time will be less. Or I’m missing something?

Good Luck!

1 Like

The thing is I am trying to make a mental health chatbot, so variety can be huge but the target emotions are depression stress anxiety guilt and anger so a sentence can have the those intents. Hence so many examples. But does rasa support it? Or should I train a model and then use the model in custom actions?

Okay will try this… sorry i am a bit new to rasa.

@Amish 1 intent 10-20 examples based on that model trained.

I’d highly recommend seeing this: Training Data Format and getting some idea on rasa intents and examples.


If you have some bot user conversation enlighten us :slight_smile: