Release koboldcpp-1.117 · LostRuins/koboldcpp

GitHub
KoboldCpp version 1.117 introduces Ollama compatibility, improved multimodal image handling, group chat reworks, and various bug fixes and performance updates.

Summary

The koboldcpp-1.117 release brings significant enhancements, most notably adding Ollama-compatible embeddings, streaming, and tool calling, allowing seamless integration with Ollama-based software. Key functional updates include inline multimodal image insertion, a reworked group chat system with improved memory management, and increased generation limits. Additionally, the release incorporates upstream fixes, updates to Kobold Lite, and minor audio improvements, while warning users that support for '--splitmode row' in CUDA will be removed in future versions.

(Source:GitHub)