Cover Image for Tiny Aya: Building the most capable multilingual open weight 3B model
Cover Image for Tiny Aya: Building the most capable multilingual open weight 3B model
Avatar for Lossfunk Event Calendar
Your friendly neighborhood AI lab
64 Went

Tiny Aya: Building the most capable multilingual open weight 3B model

Register to See Address
Bengaluru, Karnataka
Registration
Past Event
Welcome! To join the event, please register below.
About Event

In this talk, Viraat will walk through Tiny Aya, a new family of open-weight multilingual models built to be both highly capable and efficient enough to run on consumer hardware and phones.

He will cover the released models: TinyAya-Base (3.35B, 70+ languages), TinyAya-Global, and the regional specialists TinyAya Earth, TinyAya-Fire, and TinyAya-Water.⁠

⁠He will discuss the key research and product decisions behind Tiny Aya, including multilingual post-training, model specialisation, target-language quality, and balancing performance across diverse languages.⁠ ⁠

Viraat will share lessons from the training setup, evaluation design, and the tradeoffs involved in building accessible models for real researchers, developers, and communities worldwide.

Speaker

Viraat Aryabumi, Member of Technical Staff at Cohere, working on multilingual and multimodal models including Command A, Aya and Tiny Aya.

Previously built large-scale ad systems at Amazon, MSc in AI from the University of Edinburgh.

https://www.linkedin.com/in/viraataryabumi

Pre-read


Look forward to seeing you at the event.

Location
Please register to see the exact location of this event.
Bengaluru, Karnataka
Avatar for Lossfunk Event Calendar
Your friendly neighborhood AI lab
64 Went