ggerganov/llama.cpp b7960: b7960
AI Summary
The article summarizes the release of version b7960 of the ggerganov/llama.cpp project, which is a C++ implementation of the LLaMA language model. The key points are: 1. The release includes changes to reduce the number of false-positive (FA) test permutations, focusing on testing non-F16 for head sizes 64 and 72 (one a multiple of QK, one not). 2. The release supports various platforms, including macOS (Apple Silicon and Intel), iOS, Linux (Ubuntu x64 CPU and Vulkan, Ubuntu s390x CPU), Windows (x64 CPU, arm64 CPU, CUDA 12, CUDA 13, Vulkan, SYCL, HIP), and openEuler (x86 and aarch64). 3. The release is signed with GitHub's verified signature, ensuring the integrity of the code.
Original Description
<details open> tests: reduce number of FA test permutations (#19381) Only test non-F16 for head size 64 and 72 (one a multiple of QK, one not). </details> **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b7960/llama-b7960-bin-macos-arm64.tar.gz) - [macOS Intel (x64)](https://github.com/ggml-org/llama.cpp/releases/download/b7960/llama-b7960-bin-macos-x64.tar.gz) - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b7960/llama-b7960-xcframework.zip) **Linux:** - [Ubuntu x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b7960/llama-b7960-bin-ubuntu-x64.tar.gz) - [Ubuntu x64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/b7960/llama-b7960-bin-ubuntu-vulkan-x64.tar.gz) - [Ubuntu s390x (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b7960/llama-b7960-bin-ubuntu-s390x.tar.gz) **Windows:** - [Windows x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/down
Details
Discussion coming soon...