Ettin Suite: SoTA Paired Encoders and Decoders Released
Action Required
Organizations can now leverage state-of-the-art language models for a wider range of tasks, including classification, retrieval, and generation, with improved accuracy and efficiency.
AI Impact Summary
Ettin Suite introduces a new suite of state-of-the-art paired encoder-only and decoder-only models trained on identical data and recipes, achieving superior performance compared to existing models like Llama 3.2 and SmolLM2. This capability represents a significant advancement in open-data language model development, offering a controlled comparison between architectures and enabling apples-to-apples performance evaluations. The availability of diverse model sizes (17M-1B params) and a reproducible training recipe provides researchers and developers with a valuable tool for exploring the nuances of encoder and decoder architectures.
Affected Systems
- Date
- Date not specified
- Change type
- capability
- Severity
- high