The ggml-org team released llama.cpp b12419 with a server change: it now ignores empty stop strings. Stop strings are text sequences that tell a model when to end a response.
The release page lists builds for macOS, iOS and Linux, including CPU and Vulkan versions for some Linux systems.
Why it matters
People running llama.cpp’s server can supply empty stop strings without the server treating them as stop sequences; the release notes do not explain other behavior changes.