Closed
Description
As titled.
empirical.py", line 49, in _jvp
with forwardprop.ForwardAccumulator(primals, tangents) as acc:
File "/Library/Frameworks/Python.framework/Versions/3.8/lib/python3.8/site-packages/tensorflow/python/eager/forwardprop.py", line 363, in __init__
self._watch(primals, tangents)
File "/Library/Frameworks/Python.framework/Versions/3.8/lib/python3.8/site-packages/tensorflow/python/eager/forwardprop.py", line 414, in _watch
nest.map_structure(_watch, primals, tangents, expand_composites=True)
File "/Library/Frameworks/Python.framework/Versions/3.8/lib/python3.8/site-packages/tensorflow/python/util/nest.py", line 630, in map_structure
assert_same_structure(structure[0], other, check_types=check_types,
File "/Library/Frameworks/Python.framework/Versions/3.8/lib/python3.8/site-packages/tensorflow/python/util/nest.py", line 401, in assert_same_structure
raise type(e)("%s\n"
ValueError: The two structures don't have the same nested structure.
First structure: type=tuple str=([(<tf.Tensor 'stateless_random_normal:0' shape=(7, 1024) dtype=float32>, <tf.Tensor 'stateless_random_normal_1:0' shape=(1, 1024) dtype=float32>), (), (<tf.Tensor 'stateless_random_normal_2:0' shape=(1024, 1) dtype=float32>, <tf.Tensor 'stateless_random_normal_3:0' shape=(1, 1) dtype=float32>)],)
Second structure: type=tuple str=([(ndarray<<tf.Tensor 'gradient_tape/Tensordot/transpose/transpose:0' shape=(7, 1024) dtype=float32>>, ndarray<<tf.Tensor 'gradient_tape/mul_1/Mul:0' shape=(1, 1024) dtype=float32>>), (), (ndarray<<tf.Tensor 'gradient_tape/Cast_4/Cast:0' shape=(1024, 1) dtype=float32>>, ndarray<<tf.Tensor 'gradient_tape/mul_5/Mul:0' shape=(1, 1) dtype=float32>>)],)
More specifically: Substructure "type=ndarray str=ndarray<Tensor("gradient_tape/Tensordot/transpose/transpose:0", shape=(7, 1024), dtype=float32)>" is a sequence, while substructure "type=Tensor str=Tensor("stateless_random_normal:0", shape=(7, 1024), dtype=float32)" is not
Entire first structure:
([(., .), (), (., .)],)
Entire second structure:
([(., .), (), (., .)],)
Metadata
Assignees
Labels
No labels