Merge pull request #118 from Jittor/gmh

update matmul_transpose

python/jittor/nn.py

@@ -17,14 +17,14 @@ import math
 from jittor.pool import Pool, pool, AdaptiveAvgPool2d
 from jittor.optim import *
 
+
 def matmul_transpose(a, b):
     '''
     returns a * b^T
     '''
     assert len(a.shape) >= 2 and len(b.shape) == 2
     assert a.shape[-1] == b.shape[-1]
-    if jt.flags.use_cuda:
-        jt.compile_extern.cublas_ops.cublas_batched_matmul(a, b, 0, 0)
+
     shape = list(a.shape)[:-1] + list(b.shape)
     a = a.broadcast(shape, [len(shape)-2])
     b = b.broadcast(shape)