Proposal summary

Currently the evaluation metrics available in Opik are quite limited. Would be great to add additional metrics, Microsoft has a good guide on these here.

Here are some metrics we could add:

• Reference based metrics:
  • BLEU
  • ROUGE
  • JS-Divergence
• Reference-free metrics:
  • Supert
  • BLANC
  • ROUGE-C

Motivation

No response