TL;DR: llama.cpp's server now treats null sampling parameters as requests for server defaults, aligning with OpenAI's API specification.
Summary: The latest llama.cpp server update (b9967) extends schema validation to interpret null values for sampling parameters (like temperature, top_p) as requests for the server's default settings. This change allows clients to send null for nullable parameters, matching the OpenAI spec and existing json_value conventions within llama.cpp.
Why it matters: This improves API compatibility and flexibility for developers building applications on top of llama.cpp, simplifying client-side logic for parameter handling. AI builders should update their llama.cpp instances to leverage this standardized parameter behavior.
Source: github_releases