Open medical intelligence, built to run locally.
Website 路 Flagship model 路 Technical report 路 Runtime
## Featured release ### [Pestle-27B-Ternary-GGUF](https://huggingface.co/Doses-AI/Pestle-27B-Ternary-GGUF) A 27B open-weight medical model packaged as an 8.48 GB runnable GGUF. Pestle is designed for private local and on-premises systems spanning clinical knowledge, biomedical evidence work, pharmaceutical retrieval, and coding. **Apache 2.0** 路 **27B parameters** 路 **8.48 GB GGUF** 路 **CPU, Metal and CUDA** [Weights and model card](https://huggingface.co/Doses-AI/Pestle-27B-Ternary-GGUF) 路 [Technical report](https://doses.ai/pestle-27b-ternary-technical-report/) 路 [Run with mortar.cpp](https://github.com/DosesAI/mortar.cpp) ## Research - **Low-bit model compression** - ternary and compact representations for capable local models. - **Medical model evaluation** - matched benchmarking across clinical knowledge, biomedical QA, retrieval, and general capabilities. - **Local inference** - practical execution across CPU, Apple Silicon, and NVIDIA GPUs without a required cloud dependency. ## Open source | Project | What it is | Links | |:--|:--|:--| | **Pestle-27B-Ternary** | Open-weight 27B medical research model | [Model](https://huggingface.co/Doses-AI/Pestle-27B-Ternary-GGUF) 路 [Report](https://doses.ai/pestle-27b-ternary-technical-report/) | | **mortar.cpp** | Local CPU, Metal, and CUDA runtime for Pestle | [GitHub](https://github.com/DosesAI/mortar.cpp) | Research preview. Doses AI models are not medical devices and are not intended to replace qualified professional judgment. Contact: malik@doses.ai.