ggml-org has released llama.cpp v1.0.0, the project’s first 1.0 release. The changelog covers work since v0.9.0, including new hardware-backend support and performance changes, plus server security fixes and updates to the chat interface.
The release adds FP8 support for the key-value cache, an option used to store conversation context, and adds an OpenAI-compatible speech-generation endpoint to the server. The release page says more information would be available at 3 p.m. CET, but does not give a fuller explanation of what the 1.0 milestone means.
Why it matters
People running llama.cpp’s server can use its new speech-generation endpoint, though the release notes do not specify which models support it or how to configure it.
