Enable multi-model LLM evaluation and deprioritize peak shaving in grading

Deployment fix: removed binary files from history