llama.cpp

32fb9432 - Simplify GLM 4.5 parser and fix test injection for thinking_forced_open

Simplify GLM 4.5 parser and fix test injection for thinking_forced_open Parser restructure: - Follow cleaner pattern: root ::= thinking? (tools | content) - content ::= json_schema | rest (both with optional leading newline) - Unified thinking block handling for all cases - Check json_schema before tools (specific structured output first) Test fix: - When thinking_forced_open=true and injecting reasoning, include closing </think> tag - Parser expects reasoning content to end with </think> when prompt ends with <think> GLM 4.6: 11/11 needle streaming tests now pass DeepSeek R1: 7/11 (improved from 4/11, tool parsing still needs work) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>