llama.cpp

afe2fca1 - Fix needle tests for parsers using literal_tag for parameter names

Fix needle tests for parsers using literal_tag for parameter names **Problem:** Needle tests were using needle markers (e.g., $N1AK$_0$N2AK$_0) as parameter keys, but parsers that use p.literal_tag(Tag::TOOL_ARG_NAME, it.key()) expect exact parameter names from the schema (e.g., "code"). **Solution:** - Generate dynamic tool schemas where parameter names ARE the needle markers - This allows the parser's literal_tag matching to work correctly - Update Llama 3.1 template info to set tools_emit_content_with_calls=false (builtin tools don't support mixing content with tool calls) **Results:** - Llama 3.1: 4/5 passing (was 2/5) - Overall: 10/20 templates fully passing Note: Llama 3.1 parallel-tool-calls fails because template explicitly raises "This model only supports single tool-calls at once!" 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>