The llama.cpp project has released build b10005, which includes specific updates for the DeepseekV4 model architecture. The release addresses sequence management issues by fixing `seq_rm` and implementing a proper `seq_cp` function.

  • Fixes `seq_rm` for DeepseekV4 models.
  • Implements proper `seq_cp` functionality.
  • Creates an actual update context for improved handling.
  • macOS Apple Silicon (arm64) builds are available, while KleidiAI support is disabled.
  • Linux binaries cover Ubuntu x64 and arm64 (CPU), s390x, Vulkan, ROCm 7.2, OpenVINO, and SYCL FP32/FP16.
  • Android arm64 CPU builds are provided.
  • Windows releases include CPU, OpenCL Adreno, CUDA 12/13, Vulkan, OpenVINO, SYCL, and HIP variants.
  • openEuler support for x86 and aarch64 (310p/910b ACL Graph) is included, with some configurations disabled.

This update ensures correct sequence handling for DeepseekV4 models across all supported platforms.