本文整理汇总了Python中tensorflow.to_bfloat16方法的典型用法代码示例。如果您正苦于以下问题：Python tensorflow.to_bfloat16方法的具体用法？Python tensorflow.to_bfloat16怎么用？Python tensorflow.to_bfloat16使用的例子？那么恭喜您, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类tensorflow的用法示例。
在下文中一共展示了tensorflow.to_bfloat16方法的3个代码示例，这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞，您的评价将有助于系统推荐出更棒的Python代码示例。
示例1: _to_bfloat16_unbiased
# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import to_bfloat16 [as 别名]
def _to_bfloat16_unbiased(x, noise):
"""Convert a float32 to a bfloat16 using randomized roundoff.
Args:
x: A float32 Tensor.
noise: a float32 Tensor with values in [0, 1), broadcastable to tf.shape(x)
Returns:
A float32 Tensor.
"""
x_sign = tf.sign(x)
# Make sure x is positive. If it is zero, the two candidates are identical.
x = x * x_sign + 1e-30
cand1 = tf.to_bfloat16(x)
cand1_f = tf.to_float(cand1)
# This relies on the fact that for a positive bfloat16 b,
# b * 1.005 gives you the next higher bfloat16 and b*0.995 gives you the
# next lower one. Both 1.005 and 0.995 are ballpark estimation.
cand2 = tf.to_bfloat16(
tf.where(tf.greater(x, cand1_f), cand1_f * 1.005, cand1_f * 0.995))
ret = _randomized_roundoff_to_bfloat16(x, noise, cand1, cand2)
return ret * tf.to_bfloat16(x_sign) 示例2: _blocked_and_dtype_transformations
# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import to_bfloat16 [as 别名]
def _blocked_and_dtype_transformations(tensor):
"""Yields variants of a tensor, for standard blocking/dtype variants.
Args:
tensor (tf.Tensor): Input tensor.
Yields:
(modified_tensor, suffix) pairs, where `modified_tensor` is a transformed
version of the input, and `suffix` is a string like "/blocked32".
"""
for blocking_level in (32, 48):
blocked = make_padded_blocked_matrix(tensor, blocking_level)
bfloat16_blocked = tf.to_bfloat16(bfloat16_permutation(blocked))
yield blocked, '/blocked{}'.format(blocking_level)
yield bfloat16_blocked, '/blocked{}/bfloat16'.format(blocking_level) 示例3: bfloat16_permutation
# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import to_bfloat16 [as 别名]
def bfloat16_permutation(tensor):
"""Permutes values in the last dimension of a tensor.
This permutation is used so that we can directly use unpacklo/unpackhi AVX2
instructions on the matrix coefficients. These unpacking instructions
effectively permute the data. See FastUnpackPermutation() and
AvxFloatVecArray::Load(const TruncatedFloat16 *) in avx_vector_array.h for
more details.
Args:
tensor: Blocked matrix, the result of make_padded_blocked_matrix(). Must
have its last dimension a multiple of 16.
Returns:
Permuted matrix, suitable for calling tf.to_bfloat16() on. For testing
convenience we don't do so in this method.
Raises:
ValueError: If the matrix's block dimension is not a multiple of 16.
"""
orig_shape = tensor.shape
if tensor.shape[-1] % 16 != 0:
raise ValueError('Bad block dimension, must be divisible by 16')
permutation = [0, 1, 2, 3, 8, 9, 10, 11, 4, 5, 6, 7, 12, 13, 14, 15]
indices = tf.constant(
[16 * (i // 16) + permutation[i % 16] for i in xrange(orig_shape[-1])])
return tf.gather(tensor, indices, axis=len(orig_shape) - 1)