llama 3.3 70b q4_k_m - tested on yaml with lots of anchors and aliases

perplexity holds fine to 20.6k then completely breaks when you have 12+ yaml anchors with nested alias references.... the model starts generating invalid anchor names and duplicating alias blocks tested on kubernetes helm charts with heavy anchor use. failure mode is it loses track of which anchors are defined and starts referencing &nonexistent_anchor