accelerate

cff24503 - Relax numerical tolerance for XPU in test_big_modeling (#4001)