Alice AI Foundation releases 80B-A3B Base model for public use
Summary
The Alice AI Foundation has released the weights of its 80B-A3B Base model, which contains 80 billion parameters, with 3 billion being active per token, now available for community use on Hugging Face under an Apache 2.0 license. This model has been specifically tuned for strong performance in factual knowledge tasks in Russian and includes new benchmarks and evaluation protocols aimed at enhancing its capabilities. The release allows researchers and developers to experiment with the model, contributing to its ongoing improvement and application in various contexts.
Analysis
Yandex: Yandex is a leading Russian technology company with extensive work in search, AI, and machine learning infrastructure. It hosts the model repository on Hugging Face and provides detailed technical documentation, inference examples, and fine-tuning guidance for the released model. The company supports open release of pretrained weights under permissive licensing for research purposes. Alice AI Foundation: Alice AI Foundation focuses on developing advanced autoregressive language models with hybrid architectures and Mixture of Experts components. In this release, the foundation has open-sourced its latest base model to enable community experimentation and further development. The model emphasizes reasoning, tool use, and strong performance on Russian-language factual tasks. Research Focus: The release includes new Russian-oriented benchmarks and evaluation protocols to support further model improvement. Open Weights Availability: The model weights have been published on Hugging Face under the Apache 2.0 license for unrestricted community use and modification. Russian Language Emphasis: The model was specifically tuned and benchmarked for strong results on factual knowledge tasks in Russian.
Categories
aimachine_learningtech