Viraat Aryabumi
Viraat Aryabumi is a machine learning researcher whose work spans multilingual language models and code in pre-training. By 2024, he had earned a bachelor's degree in information technology at CBIT, Osmania University, and a master's in artificial intelligence at the University of Edinburgh; he had also worked in display advertising at Amazon, led machine learning and AI at Aiara, and joined Cohere For AI as a Research Scholar. In February 2024, he was second author of the Aya Model paper; in May 2024, he was first author of Aya 23 and presented QLoRA fine-tuning at DataHack Summit 2024. He submitted a code-pretraining paper in August 2024 that appeared at ICLR 2025, coauthored Aya Expanse in December 2024, contributed to Command A in April 2025, and coauthored the Tiny Aya report in March 2026. Since 2025, he is a Member of Technical Staff at Cohere.
- Now at
- Cohere
Timeline
-
By 2024
Degrees and AI roles
He earns a bachelor's degree in engineering, majoring in information technology, from CBIT, Osmania University, and a master's in artificial intelligence from the University of Edinburgh's School of Informatics. He works on display advertising as a machine learning scientist at Amazon Development Centre in Scotland, then heads machine learning and AI at Aiara, an AI-for-manufacturing startup. By 2024, he is a Research Scholar at Cohere For AI.
-
February 2024
The Aya Model paper
In February 2024, he is second author of the Aya Model paper, an instruction-finetuned generative language model that follows instructions in 101 languages. He helps develop the Aya Project and release the Aya-101 model.
-
May 2024
Aya 23 open-weight models
In May 2024, he is first author of the Aya 23 technical report, which introduces multilingual models serving 23 languages and releases open weights for 8B and 35B models. At DataHack Summit 2024, he presents fine-tuning an open-weights aya-23-8B model on a new language with QLoRA.
-
2024–2025
Code in pre-training data
In August 2024, he is first author of “To Code, or Not To Code? Exploring Impact of Code in Pre-training,” with Yixuan Su, Raymond Ma, Adrien Morisot, Ivan Zhang, Acyr Locatelli, Marzieh Fadaee, Ahmet Üstün and Sara Hooker. In 2025, the paper appears at the International Conference on Learning Representations; its experiments report up to an 8.2% relative gain in natural-language reasoning and a 12x boost in code performance compared with text-only pre-training.
-
December 2024
Aya Expanse models
In December 2024, he is one of Aya Expanse’s 45 authors. The report introduces 8B and 32B multilingual models that outperform Gemma 2, Qwen 2.5 and Llama 3.1 in their parameter classes.
-
April 2025
Command A for enterprise
In April 2025, he is among Team Cohere’s authors of Command A, an agent-optimized enterprise model that supports 23 languages.
-
March 2026
Tiny Aya for 70 languages
In March 2026, he coauthors the Tiny Aya report, which describes a multilingual model trained on 70 languages and refined through region-aware posttraining, with 3.35B parameters.
-
Since 2025
Member of Technical Staff
Since 2025, he is a Member of Technical Staff at Cohere and a speaker at the Cohere Labs Connect Conference.
Career
- Since 2025CohereResearch Scholar
- AmazonMachine Learning Scientist
- AiaraHead of Machine Learning and AI
Education
- Osmania UniversityBachelor's in Engineering in information technology
- University of EdinburghMaster's in Artificial Intelligence
Sources and credits
Photo: We have no published photo of Viraat Aryabumi yet. Send us a free-licensed photo.
- Viraat Aryabumi
- Analytics Vidhya | DataHack Summit 2024
- [2402.07827] Aya Model: An Instruction Finetuned Open-Access Multilingual Language Model
- [2405.15032] Aya 23: Open Weight Releases to Further Multilingual Progress
- [2408.10914] To Code, or Not To Code? Exploring Impact of Code in Pre-training
- To Code or Not to Code? Exploring Impact of Code in Pre-Training | ML Anthology
- [2412.04261] Aya Expanse: Combining Research Breakthroughs for a New Multilingual Frontier
- [2504.00698] Command A: An Enterprise-Ready Large Language Model
- [2603.11510] Tiny Aya: Bridging Scale and Multilingual Depth
- Open Science Community - Connect Conference

