llama.cpp
217b8a3a - Fix parallel tool calls in multiple parsers

Commit
101 days ago
Fix parallel tool calls in multiple parsers MiniMax M2, Seed OSS: - Default allow_additional to true per JSON Schema spec - Add trailing p.space() after tool_close to consume whitespace between calls Functionary V3.1: - Remove overly strict python tool validation that rejected multi-arg tools - Add needle_tool_name config to avoid triggering special python handling in tests Qwen3 Coder (partial): - Default allow_additional to true - Add <tool_call>/</ tool_call> wrapper matching - Use p.space() instead of literal spaces for flexibility Note: Tool call scenarios still failing - needs deeper investigation Results: 18/20 templates pass (was 15/20) - GLM 4.6: 9/9 ✓ - MiniMax M2: 7/7 ✓ (was 6/7) - Seed OSS: 9/9 ✓ (was 8/9) - Functionary V3.1: 5/5 ✓ (was 2/5) - Qwen3 Coder: 2/5 (unchanged, complex format) - Granite: 4/7 (pre-existing) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Author
Parents
Loading