would the examples used to generate new example mixed with different intents?

ServiceNow / data-augmentation-with-llms

Data Augmentation for Intent Classification with Off-the-Shelf Large Language Models is a ServiceNow Research project

Apache License 2.0

28 stars 0 forks source link

would the examples used to generate new example mixed with different intents? #3

Open troilus-canva opened 1 year ago

troilus-canva commented 1 year ago

https://github.com/ServiceNow/data-augmentation-with-llms/blob/77c740df2aaf4bfc1b2449114613f98b37a1faae/utils/data_utils/augment_slices.py#L204

Demfier commented 1 year ago

Hi @troilus-canva,

So sorry for this late reply. If you haven't already found the answer: no, the generated examples would never appear in the prompt. That block of code simply takes the first k examples from the training set (and we ensure that the generated examples are always appended at the end)