mamba2

llama.cpp 0.3.0 introduces multimodal capabilities with the dots3-note model and MTP support for GLM-4.5-Air, alongside significant updates to ggml and core model handling.

Optimized mamba2 by flattening projections for GEMM dispatch and removed redundant reshape operations. This release also provides pre-compiled binaries for various platforms and hardware accelerators.

This release focuses on improving Mamba2 model conversion by removing the hardcoded 2x expansion factor and fixing an erroneous parameter check. It also provides updated pre-built binaries for broad platform compatibility.

This release includes a fix for an assertion in mamba2 and introduces the ability to add n_group modifications.