// blog · 2 posts
blog
Benchmarks, releases and the experiments that did not work, for the languages of Nepal. Every claim links to a repo, a model card or a dataset.
· release · raniNepaliConformer is now Rani
Rani continues NepaliConformer with TDT and CTC heads and a Nepali vocabulary. This report records its training stages, failed gates, public-system comparisons, phone-call gains...
· release · yak tokenizerYak: An Open, Lossless, Nepali-First Tokenizer
An open, lossless, Nepali-first tokenizer. Its phrase build is #1 lossless on Nepali of the 42 we measured, and v2.0 is the lowest-cost lossless tokenizer on 18 of 19 languages...