What's new? |
Turning on Tensorflow XLA - 컴퓨터 |
Version: f1f85733439920c031519042e312d6bbe60f5f93
TF_PATH=`pip show tensorflow | grep Location | cut -f2 -d" "`
export TF_XLA_FLAGS="--tf_xla_enable_xla_devices=true --tf_xla_auto_jit=2 --tf_xla_cpu_global_jit" $TF_PATH
TF_PATH=`pip show tensorflow | grep Location | cut -f2 -d" "`
export TF_XLA_FLAGS="--tf_xla_enable_xla_devices=true --tf_xla_auto_jit=2 --tf_xla_cpu_global_jit" $TF_PATH
written time : 2020-09-13 14:46:16.0
Tensorflow 2.4.0 - 컴퓨터 |
checking whether mkl is enabled or not.
python -c "from tensorflow.python.framework import test_util;print(test_util.IsMklEnabled())"
python -c "from tensorflow.python.framework import test_util;print(test_util.IsMklEnabled())"
written time : 2020-09-10 15:31:35.0
PyTorch vs Tensorflow - 컴퓨터 |
Unexpected differences
>>> a=np.random.rand(500,500)
>>> b=tf.pow(tf.constant(a, tf.float32),2)
>>> c=torch.pow(torch.tensor(a, dtype=torch.float32),2)
>>> np.sum(b.numpy()-c.numpy())
3.87448e-06
>>> a=np.random.rand(500,500)
>>> b=tf.pow(tf.constant(a, tf.float32),2)
>>> c=torch.pow(torch.tensor(a, dtype=torch.float32),2)
>>> np.sum(b.numpy()-c.numpy())
3.87448e-06
written time : 2020-09-08 12:25:36.0