instruction for other use cases like intent classification
#5
by
SudipThomas
- opened
What should be the instructions for use cases like intent classification where we have to find embeddings of intent name, intent description, intent example sentences and the unknown sentence for which the intent needs to be found out using cosine similarity?
Are there specific instructions which were used to the train the model for different tasks?
The instruction is used only when we need to find a long description for a short sentence. And we only have one instruction for the short sentence: "Represent this sentence for searching relevant passages: ". Any other instruction is not used in training. For intent classification, there is no need to add instruction.