feat: add benchmark.py (Qwen3 vs baseline) and multi-model eval script

922932a

Surya-sj commited on Apr 26