llama.cpp merges token-aware parsing PR for more reliable tool calling and higher t/s

jacek2023 · reddit · 2026-11-23

llama.cpp merged PR #29307 'chat: implement token aware parsing' by aldehir — 62 commits, +1988/-502 lines, 21 comments and 5 reviews. The change makes chat-side parsing token-boundary aware, aiming for more reliable tool calling and higher tokens/second. A substantive improvement for anyone running local LLMs with function calling / agent workflows.

Original post →

More from coding & agent

coding & agent channel →