ShinkaEvolve is an innovative evolutionary code optimization framework that utilizes large language models (LLMs) to discover new algorithms with unprecedented sample efficiency. It has achieved state-of-the-art solutions in various domains, including Circle Packing and agent design, by significantly reducing the number of samples needed for effective program evolution. The framework is open-sourced to empower researchers and engineers in their scientific discoveries and development efforts.

Character.AI has open-sourced pipeling-sft, a scalable framework designed for fine-tuning large-scale MoE LLMs like DeepSeek V3. This framework addresses challenges in training efficiency and stability, integrating multi-level parallelism and supporting various precision formats, while facilitating seamless HuggingFace integration for researchers.