Add scripts/reproduce.py + benchmark_latency.py; pin reqs to working transformers/torch range; refresh README with adapter overhead numbers

8e4e948

verified

RiverRider commited on 16 days ago