Open Source
llama.cpp b11153 adds ROCm 10 and Snapdragon support
llama.cpp Releaseshigh confidence
Why it matters
- →ROCm 10.0 support removes barriers for AMD GPU users running local inference on Linux and Windows.
- →New Snapdragon arm64 builds enable efficient AI execution on mobile hardware using Hexagon NPU acceleration.
- →CUDA 13.4 integration ensures compatibility with the latest NVIDIA driver stacks across multiple architectures.
llama.cpp has released build b11153, expanding hardware support to include ROCm 10.0 for AMD GPUs on Linux and Windows, as well as native arm64 builds for Qualcomm Snapdragon processors on Linux. The update also adds CUDA 13.4 libraries for both x64 and arm64 architectures across Ubuntu and Windows platforms. Conversely, KleidiAI support on macOS Apple Silicon has been disabled in this release, and openEuler builds are currently marked as disabled. This release underscores the project's strategy of maintaining broad compatibility across diverse hardware ecosystems.
Read originalMore from llama.cpp Releases
Coding Toolscoding
llama.cpp b11140 optimizes sparse attention on CUDA
This release targets a specific bottleneck in long-context inference by optimizing the sparse flash attention prefill step for NVIDIA GPUs. By templating kernels to unroll loops at compile time, batched sparse operations drop from 586 microseconds to 244 microseconds on 49k context windows. This isn't just a generic speed bump; it makes handling very long documents significantly more efficient for users relying on sparse attention mechanisms. The change is already baked into the standard CUDA builds, requiring no special flags.
llama.cpp Releases
Coding Toolscoding
llama.cpp b11146 adds CUDA 13 and Snapdragon support
The latest llama.cpp build brings immediate relevance to users on bleeding-edge NVIDIA hardware with native CUDA 13.4 support across Linux and Windows, closing the gap for those testing next-gen GPU architectures. More notably, it finally addresses the mobile inference landscape by including a dedicated build for Linux arm64 Snapdragon devices, covering CPU, Adreno GPU, and Hexagon NPU paths. This moves local AI beyond just desktop GPUs into the realm of high-performance edge computing on Qualcomm silicon. While Apple Silicon builds have KleidiAI disabled in this specific release, the expansion to ARM-based mobile NPUs marks a significant shift in where llama.cpp can run efficiently.
llama.cpp Releases
Coding Toolscoding
llama.cpp b11147 adds ROCm 10 and Snapdragon support
This release quietly cements llama.cpp as the universal inference runtime by finally bringing first-class ROCm 10.0 support to both Linux and Windows. AMD GPU users no longer need workarounds, effectively closing a long-standing parity gap with NVIDIA's CUDA ecosystem. Equally notable is the new Linux arm64 build targeting Snapdragon chips, which unlocks local LLM execution on high-performance mobile hardware via CPU, Adreno GPU, and Hexagon NPU acceleration. While KleidiAI on Apple Silicon has been disabled in this specific binary set, the broader platform expansion signals a shift toward heterogeneous computing that extends well beyond traditional desktop GPUs.
llama.cpp Releases
More in Open Source
© Together AI BlogOpen Sourcecoding
Exploring the Open Source AI Stack
The open source AI stack is becoming increasingly attractive to developers and organizations seeking more control and cost efficiency compared to closed-source models. This stack, known as the MIGHT Stack, consists of independent layers including models, inference, gateways, harnesses, and tools, allowing for flexible and customizable development workflows. Large models like Kimi K3 offer robust capabilities for complex tasks, while smaller models like GLM 5.3 Flash provide cost-effective solutions for well-defined tasks. This modular approach enables developers to experiment with new models quickly, adapting to the fast-paced evolution of AI technology.
Together AI Blog
© Cole MedinOpen Sourcecoding
AI Software Factory: Open Source Alpha Released
Cole Medin is pioneering a new approach to AI-driven software development with his AI Software Factory. This open-source project automates the entire coding process from a product requirements document to shipped code, without human intervention in the coding phase. Medin's experiment has already produced Dynachat, an AI tutor, demonstrating the potential of this method. Now, he's inviting developers to explore and contribute to the early alpha version, aiming to make this transformative tool accessible to all. This could redefine how software is developed, offering a glimpse into a future where AI handles the heavy lifting of coding.
Cole Medin
© Lev SelectorOpen Sourcemodels
DeepSeek Harness Gains 200,000 GitHub Stars in a Week
DeepSeek Harness has rapidly gained popularity, reaching nearly 200,000 stars on GitHub within a week of its release.
Lev Selector