llama.cppProduct update
llama.cpp · v0.4.0
llama.cpp 0.4.0 adds initial support for the Qwen3.8-Flash-Next architecture, with optimization work still pending, as well as NVIDIA Nemotron-3-Puzzle-75B-A9B. The core can read tensors on demand with --lazy-mode, the server can cap context per slot, and video input options arrive. The bundled ggml library moves from 0.22.0 to 0.23.0 with sparse flash attention, RPC event and async APIs, and Apple RDMA transport. The server enables preserve_reasoning by default and rejects prefilled assistant tool calls.
Saved . A shortened excerpt from GitHub Releases; the original publisher retains authorship.