verify5.py 439 B

123456789
  1. import time
  2. t = time.time()
  3. from flashinfer.jit.gemm import gen_gemm_sm120_module_cutlass_mxfp8 as g1
  4. r = g1().try_load()
  5. print("VERIFY mxfp8: %s %.1fs" % ("HIT" if r is not None else "MISS", time.time() - t), flush=True)
  6. t = time.time()
  7. from flashinfer.mla._sparse_mla_sm120 import get_sparse_mla_sm120_module as g2
  8. g2(); dt = time.time() - t
  9. print("VERIFY sparse_mla: %s %.1fs" % ("HIT" if dt < 15 else "MISS-COMPILED", dt), flush=True)