ggerganov/llama.cpp Released Version b11828
llama.cpp release b11828 updates the server's realtime audio API to correctly handle task completion once committed audio streams have fully drained. This fix ensures that audio data appended during tail draining and back-to-back commits are properly managed without leaking or merging incorrectly. This update improves the reliability and precision of realtime audio processing within the llama.cpp server ecosystem. Developers building local voice or audio-driven applications will experience more robust session management and fewer synchronization bugs. The update introduces mtmd_audio_stream_flush() to record row counts at the end of flushed audio, using a FIFO slot queue to complete each commit strictly after every preceding row is decoded.
## BACKGROUND
llama.cpp is a widely used open-source library that enables efficient execution and inference of large language models on consumer-grade hardware. It includes a built-in server component that exposes various APIs, including endpoints designed for streaming and real-time audio interactions.