Closed SuryaKrishna02 closed 3 months ago
Hi! I will be making a PR soon, in the meantime I wanted to discuss methodology. I want to add a few different methods, one of which would be to use an LLM as a judge to verify conciseness, consistency, etc on the translation. @Asnegha and I will also include statistical methods to flag problematic translations. Does this sound alright?
@asusevski @Asnegha Sounds Great. Thanks for taking up work. I guess it is better to connect and discuss these over a call so that we can scope what we can do for Aya Expedition. Or else you can mention here the brief steps/methodologies you are planning to do.
Tasks to be completed by over the next week to close out this issue:
Develop a Verification Script to verify the translations of the generated dataset from the
c4ai-aya-23
model.Please feel free to add more thoughts on this to better find out the faulty translations from the model.