Models & Labs
BF16 Support Added to llama.cpp SYCL Backend
llama.cpp Releaseshigh confidence
Why it matters
- →Adds BF16 support to prevent inefficient GPU-to-CPU transfers.
- →Enhances performance for models using BF16 embedding tensors.
- →Aligns with existing data type handling, improving efficiency.
The b9088 release of llama.cpp introduces BF16 support to the SYCL backend's GET_ROWS operation, resolving a performance regression issue. This update prevents models using BF16 embedding tensors from defaulting to CPU processing, which previously caused inefficient GPU-to-CPU tensor transfers. By leveraging the existing template for sycl::ext::oneapi::bfloat16, the update aligns with the handling of other data types, enhancing performance across supported platforms. This improvement is crucial for developers utilizing BF16 models, ensuring more efficient and streamlined operations.
Read originalMore from llama.cpp Releases
Models & Labsmodels
Llama.cpp adds GLM-5.2 speculative decoding support
Llama.cpp's latest update introduces speculative decoding support for GLM-5.2, enhancing its capabilities with NextN/MTP features. This addition allows for more efficient tensor loading and context management, particularly benefiting models using the GLM_DSA architecture. The update also includes options for exporting models with or without the MTP feature, providing flexibility for developers. This release marks a step forward in optimizing model performance and adaptability, especially for those leveraging the GLM-5.2 framework.
llama.cpp Releases
Open Sourcemodels
llama.cpp b10175 Release Expands Platform Support
The latest b10175 release of llama.cpp continues its trend of broadening platform compatibility, making it a versatile tool for developers across different systems. Notably, this update includes support for ROCm 7.2 on Ubuntu x64, which is significant for AMD GPU users seeking alternatives to NVIDIA's CUDA. The release also maintains a wide array of builds for Windows, macOS, and Linux, ensuring that developers can leverage llama.cpp's capabilities regardless of their hardware setup. While there are no groundbreaking new features, the consistent expansion of platform support solidifies llama.cpp's position as a flexible inference runtime option.
