Palmyra-mini family release: 1.5B–1.7B models with CoT reasoning and quantizations

AI Impact Summary

The Palmyra-mini family introduces three 1.5B–1.7B parameter models with Chain-of-Thought reasoning to boost complex tasks while preserving lightweight inference. The release includes GGUF and MLX quantizations and supports Qwen-architecture base models, with palmyra-mini-thinking-a and palmyra-mini-thinking-b targeting advanced reasoning and problem solving, respectively. Note that RL fine-tuning increases single-shot accuracy but reduces sampling diversity, so teams should benchmark these variants against the non-reasoning base when choosing for production tasks.