Muga is the institutional core of the Navdyut family: trained on the scanned Assamese corpus behind our research pipeline, using the shared 32K Assamese tokenizer, and released for on-prem and sovereign cloud installs.
Deep product docs stay on the Models page. This note records the release itself: what moved on our internal Assamese evaluation suite between Eri-scale experiments and the 30B cut.
Deep model documentation stays on the Models page. This post is the release record.