@nicolay
Treat finetuning as an experiment! It's cheap to train multiple models, so evaluate them on your own test sets & pick the best. Start with fast evals (LLM as a judge,…) before investing in production-grade testing.
0 reply
0 recast
0 reaction