llama.cpp Release b11246 Fixes JSON Schema Output for Muse Glimmer
llama.cpp release b11246 resolves a bug where the Muse Glimmer model ignored structured JSON schema response formatting when evaluated with Jinja chat templates. The patch also improves chat template handling by supporting JSON code fences and cleaning up choice formatting. Structured outputs are essential for local agentic models like Muse Glimmer to reliably produce machine-readable JSON for function calling and tool execution. Ensuring Jinja chat templates properly enforce schema formatting prevents integration failures in automated workflows. The fix was implemented in PR #29615 by contributor Alde Rojas to address issue #29613. Updated pre-built binary packages were released simultaneously across macOS, Linux, Windows, Android, and Snapdragon backends.
## BACKGROUND
llama.cpp is a popular open-source C/C++ inference framework designed to run large language models efficiently on local consumer hardware. Modern models use Jinja chat templates to structure conversational prompts, alongside JSON schema constraints to guarantee that the LLM generates valid structured data.