~/LLAMA CPP/ggerganov-llama-cpp-released-b12229-with-json-schema-fixes

ggerganov/llama.cpp Released b12229 with JSON Schema Fixes

llama.cpp release b12229 fixes a bug where top-level JSON schemas in response_format were ignored by the OpenAI-compatible server endpoint. It now correctly accepts schemas provided directly as response_format.schema in addition to the nested wrapper form. This fix ensures better compatibility with various client applications relying on the OpenAI API standard for structured outputs when running local models. It prevents silent failures where text generation previously ran unconstrained despite a requested JSON schema. The update improves how the server checks the presence of json_schema keys, giving the wrapper precedence while properly handling top-level schemas. It also adds test cases to prevent regressions and updates the documentation in the README.

## BACKGROUND

llama.cpp is an open-source software library that performs inference on various large language models and includes a server with an OpenAI-compatible API endpoint. Structured outputs in this API allow developers to enforce a strict JSON schema on the model's responses.

## REFERENCES

## KEYWORDS

#llama.cpp#llm#ai-infrastructure#github-releases

$ subscribe --daily