def testBatchNormalizeFused(self):
with self.session() as sess:
a = array_ops.placeholder(np.float32, [4, 64, 64, 4],
name="input_a")
def my_graph(a):
with ops.device("/device:IPU:0"):
with variable_scope.variable_scope("", use_resource=True):
beta = variable_scope.get_variable(
"x",
dtype=np.float32,
shape=[4],
initializer=init_ops.constant_initializer(0.0))
gamma = variable_scope.get_variable(
"y",
dtype=np.float32,
shape=[4],
initializer=init_ops.constant_initializer(1.0))
b_mean, b_var = nn.moments(a, [0, 1, 2],
name='moments')
normed = nn.fused_batch_norm(a,
gamma,
beta,
b_mean,
b_var,
is_training=False)
return normed
report = ReportJSON(self, sess)
out = ipu.ipu_compiler.compile(my_graph, [a])
sess.run(variables.global_variables_initializer())
report.reset()
result, _, _ = sess.run(out, {a: np.zeros([4, 64, 64, 4])})
self.assertAllClose(result, np.zeros([4, 64, 64, 4]))
report.parse_log()
bl = ['*convert*/Cast*']
report.assert_compute_sets_not_in_blacklist(bl)
report.assert_tensor_input_names("input_a", "x", "y")
def testBatchNormalizeLayerFusedFp16(self):
with self.session() as sess:
with ops.device("/device:IPU:0"):
with variable_scope.variable_scope("", use_resource=True):
a = array_ops.placeholder(np.float16, [4, 64, 64, 4],
name="input_a")
normed = layers_norm.batch_normalization(a, fused=True)
report = ReportJSON(self, sess)
sess.run(variables.global_variables_initializer())
report.reset()
result = sess.run(normed, {a: np.zeros([4, 64, 64, 4])})
self.assertAllClose(result, np.zeros([4, 64, 64, 4]))
report.parse_log()
bl = ['*convert*/Cast*']
report.assert_compute_sets_not_in_blacklist(bl)
report.assert_tensor_input_names("input_a")
