train CIFAR failed

[madoka109]

I use ./cifar_trainval.sh to train，but it occurs some problems，why?

Traceback (most recent call last):
File "/environment/python/versions/miniconda3-4.7.12/envs/tf1.15/lib/python3.7/site-packages/tensorflow_core/python/client/session.py", line 1365, in _do_call
return fn(*args)
File "/environment/python/versions/miniconda3-4.7.12/envs/tf1.15/lib/python3.7/site-packages/tensorflow_core/python/client/session.py", line 1350, in _run_fn
target_list, run_metadata)
File "/environment/python/versions/miniconda3-4.7.12/envs/tf1.15/lib/python3.7/site-packages/tensorflow_core/python/client/session.py", line 1443, in _call_tf_sessionrun
run_metadata)
tensorflow.python.framework.errors_impl.InternalError: 2 root error(s) found.
(0) Internal: Blas GEMM launch failed : a.shape=(128, 64), b.shape=(64, 10), m=128, n=10, k=64
[[{{node resnet/tower_0/fully_connected/dense/MatMul}}]]
[[resnet/tower_0/softmax_cross_entropy_loss/assert_broadcastable/is_valid_shape/has_valid_nonscalar_shape/has_invalid_dims/concat/_1549]]
(1) Internal: Blas GEMM launch failed : a.shape=(128, 64), b.shape=(64, 10), m=128, n=10, k=64
[[{{node resnet/tower_0/fully_connected/dense/MatMul}}]]
0 successful operations.
0 derived errors ignored.