llama.cpp Released Version b12090 With zDNN Fix
Release b12090 of llama.cpp introduces a targeted fix for the ggml-zdnn flash attention test failure under pull request #31583. This update continues routine maintenance and stabilization for specific backend integrations. This point release ensures proper functionality and testing for specialized hardware environments using IBM zDNN acceleration. It helps maintain code quality and reliability across the diverse architectures supported by the llama.cpp ecosystem. The update specifically resolves an issue within the ggml-zdnn backend implementation concerning flash attention (fattn) testing. Binaries for multiple platforms and accelerators are bundled alongside this tag.
## BACKGROUND
llama.cpp is an open-source library that enables efficient inference for large language models on consumer and enterprise hardware. The ggml-zdnn backend provides integration for IBM mainframe systems utilizing the zDNN hardware acceleration library.