Eikos - OSS Jev-like model (github.com)

The recently announced Eikos models, developed by Caio Vicentino, represent a significant innovation in open-source AI decision-making tools tailored for global finance and trading. Released under the MIT license, Eikos offers calibrated single-pass typed-decision models, available in two sizes (Eikos-4B and Eikos-27B), designed to answer structured questions about various financial scenarios. Each model processes up to 100 options in a single forward pass, providing users with confidence-inspiring probability outputs to guide their decisions based on pre-defined rules and regulations. Eikos stands out due to its versatility, offering builds optimized for bf16, FP8, INT4, and MLX for Apple Silicon, catering to a range of hardware infrastructures. The repository includes extensive resources, from serving and inference code to data pipelines and evaluation frameworks. Its ability to contextualize requests while adhering to complex rule-based systems marks a notable step in automation within finance, potentially improving decision accuracy and efficiency. The models also maintain a focus on transparency and reproducibility, with detailed documentation supporting developers and researchers in leveraging this cutting-edge technology for financial applications.