The llama.cpp project has released an update, b11454, which introduces support for the K2 Horizon dense model and its MoVA variant. This update includes significant changes to model loading, handling of hyperparameters and tensors, and adjustments to the compute graph. Additionally, the release addresses Unicode handling issues, particularly with zero-width non-joiner and joiner characters, which previously caused loading failures on Windows. The update also incorporates support for YaRN betas and enforces response schemas for reasoning and tool calls. AI
IMPACT Enhances the capabilities of local LLM inference engines by supporting new model architectures.
RANK_REASON This is a software release for an open-source project that enhances model support, fitting the research/infrastructure category.
- bitalov
- ggml-org
- K2 Horizon
- llama.cpp
- MOVA
- Codex
- CUDA
- GGUF
- Llama 3
- Natani L. Mayday
- Ryandito Diandaru
- Visual C++
- WestWaters
- YaRN
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →