b9061
New release adds long-seq tests and fixes for gated_delta_net in llama.cpp
Deep Dive
The llama.cpp b9061 release adds long-sequence test cases and fixes input ranges for gated_delta_net. Builds are available for macOS (Apple Silicon with KleidiAI, Intel x64), Linux (Ubuntu x64, arm64, s390x with Vulkan, ROCm, OpenVINO, SYCL), Windows (x64 CPU, arm64 CPU, x64 CUDA, Vulkan, SYCL, HIP), Android (arm64), and openEuler (x86, aarch64).
Key Points
- Adds long-sequence and long-sequence tail test cases for gated_delta_net
- Fixes input ranges for gated_delta_net to be more realistic
- Available for macOS, Linux, Windows, Android, and openEuler with GPU acceleration options
Why It Matters
Improves local LLM inference reliability for long contexts, enabling developers to deploy larger models with extended memory.