b8760

The llama.cpp project, a cornerstone of the open-source AI ecosystem for efficient model inference, has pushed a significant new release tagged b8760. This commit, signed with GitHub's verified signature, addresses a specific technical bug related to tensor parallelism (TP) when running the newly released Qwen 3 Next model from Alibaba. The fix corrects a data split issue that could have caused incorrect model outputs or performance degradation, ensuring stable and accurate inference for users experimenting with this advanced model.

Beyond the core fix, the release is notable for its extensive cross-platform support. The development team has provided 27 distinct pre-built binary assets, dramatically simplifying deployment. This includes optimized builds for Apple Silicon (with optional KleidiAI acceleration), Intel Macs, various Linux configurations supporting CPU, Vulkan, ROCm 7.2, and OpenVINO backends, and comprehensive Windows packages for CUDA 12.4, CUDA 13.1, Vulkan, SYCL, and HIP. Specialized builds for Huawei's openEuler OS and Ascend AI processors (310P, 910B) are also included, highlighting the project's commitment to hardware-agnostic accessibility.

This release underscores llama.cpp's role as critical infrastructure, rapidly integrating support for cutting-edge models like Qwen 3 Next and ensuring they run optimally everywhere—from a developer's laptop to specialized data center hardware. The breadth of pre-compiled binaries removes a major barrier to entry, allowing researchers and engineers to focus on building applications rather than wrestling with complex compilation toolchains for different systems.