SITUATION DETECTED: The city of Rio de Janerio has post-trained a model.
A new model named Rio 3.5 Open 397B has been developed, building upon the Qwen 7/2 architecture. This model incorporates SwiReasoning, a novel framework that optimizes token efficiency by selectively engaging in explicit "thinking out loud" processes only when necessary, while otherwise utilizing silent, latent-space reasoning. AI
IMPACT Introduces a new method for improving LLM token efficiency through dynamic reasoning, potentially impacting future model development.