Building a Large Japanese Web Corpus for Large Language Models arxiv.org 76 points · PaulHoule · 20 days ago
There is also Swallow, which is a Mistral fine tune.https://tokyotech-llm.github.io/swallow-mistralEdit: oops, this is the same team! Glad they are publishing more about it.
I find ChatGPT (4) to have excellent Japanese skills already. It seems to produce more accurate and certainly more idiomatic translations than Google Translate does, and can even explain its own translations afterwards.
xrd ·20 days ago
https://tokyotech-llm.github.io/swallow-mistral
Edit: oops, this is the same team! Glad they are publishing more about it.
LeoPanthera ·20 days ago
Show replies