projects / platform / upstream / tensorflow.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 30fcdec) | patch
[XLA] Change the TF2XLA bridge to perform F16 reduction using F32 data type.
authorBixia Zheng <bixia@google.com>
Mon, 30 Apr 2018 23:11:38 +0000 (16:11 -0700)
committerTensorFlower Gardener <gardener@tensorflow.org>
Mon, 30 Apr 2018 23:14:39 +0000 (16:14 -0700)
commit18343616da47a9c3eab79b5028ac3d8bf786f2ff
tree69fad84783fef47790c49416b2e8b5f2897d52e0tree | snapshot
parent30fcdecc05e6b25ab8d451997904e40b2a76acd4commit | diff
[XLA] Change the TF2XLA bridge to perform F16 reduction using F32 data type.

Add test cases to test that reduce sum for bfloat16 and float16 doesn't lose too
much precision.

PiperOrigin-RevId: 194862078
tensorflow/compiler/tests/reduce_ops_test.py diff | blob | history
tensorflow/compiler/tf2xla/xla_helpers.cc diff | blob | history
Domain: Machine Learning / ML Framework;