llama.cpp
5bbe7e48 - Fix Granite parser: add tool_calls formatting to template

Commit
100 days ago
Fix Granite parser: add tool_calls formatting to template The Granite Jinja template was missing logic to format tool_calls in assistant messages. This caused the systematic_needle_streaming tests to fail (4/7 passed). Changes: - Add tool_calls formatting with <|tool_call|>[...] format to template - Set tool_required_allows_content=false since Granite's system prompt says "respond only with <|tool_call|> followed by a JSON list" - Rename template with llama-cpp- prefix to indicate modification Result: Granite now passes 8/8 systematic_needle_streaming tests. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Author
Parents
Loading