A naukri.com initiative
VentureBeat
4w
337
Image Credit: VentureBeat
Cohere’s first vision model Aya Vision is here with broad, multilingual understanding and open weights — but there’s a catch
- Canadian AI startup Cohere, targeted at enterprises, releases its first vision model Aya Vision, supporting inputs in 23 languages and integrating language and vision capabilities.
- Aya Vision enhances AI's image interpretation, text generation, and multilingual translation, aiding organizations operating globally with diverse language preferences.
- Available under CC BY-NC 4.0 license, Aya Vision is accessible on Cohere's website, Hugging Face, Kaggle, and WhatsApp.
- Aya Vision enables image caption generation, visual question answering, and multilingual text-based tasks across a range of languages from English and Spanish to Arabic and Chinese.
- Despite its non-commercial licensing, Aya Vision excels in performance efficiency compared to larger models, achieving high win rates in multilingual image understanding tasks.
- Innovations like synthetic annotations, multilingual data scaling, and multimodal model merging contribute to Aya Vision's superior processing accuracy and multilingual capabilities.
- Usage of Aya Vision is suitable for AI researchers, data scientists, and enterprises for internal research, prototyping, benchmarking, and exploring AI capabilities before deployment.
- Cohere's Aya Vision represents a progressive move towards inclusive and accessible multilingual AI research, challenging larger closed-source models in the field.
- Released as part of the wider Aya initiative by Cohere, Aya Vision aims to support multilingual AI advancements with open weights available for global researchers and developers.
- Cohere's commitment to open science extends to the AyaVisionBenchmark, an evaluation set for multilingual vision, emphasizing the importance of rigorous assessment in multimodal AI development.
Read Full Article
20 Likes
For uninterrupted reading, download the app