Hi there,
When fine-tuning OpenAI ada on our text classification task (with max_tokens = 1), some of the output classes are not a complete word (example ‘Satisfaction’ → ‘Satis’).
I think this is due to OpenAI automatically cutting off the word into different tokens, as setting max_tokens = 2 do return ‘Satis’ and ‘faction’.
How can I ensure that OpenAI returns the output class as a single complete word? Thank you in advance.