Release of Ornith 1.5 LLM Family in 9B, 35B, and 397B Sizes
The Ornith 1.5 large language model family has been released on Hugging Face, featuring model sizes of 9B, 35B, and 397B parameters. These models are also available in the GGUF format, which is optimized for local inference. This release provides the open-source AI community with new options for local deployment across various hardware capabilities, ranging from lightweight edge devices to massive Mixture-of-Experts setups. It offers alternative models tailored for agentic coding and self-improving tasks. The release includes the Ornith-1.5-35B-A3B model, which likely utilizes active parameters in a Mixture-of-Experts (MoE) architecture, alongside GGUF versions for all sizes. However, detailed benchmark results and extensive documentation for this specific 1.5 version are currently limited.
## BACKGROUND
Ornith is a family of open-source, self-improving language models designed specifically for agentic coding tasks, originally built on top of pretrained models like Gemma and Qwen. The GGUF format is a binary file format designed by the llama.cpp project to store tensors and metadata efficiently, allowing users to run large models on consumer-grade hardware.