onnxruntime/include
Sherlock b03fb82ab7
Transformer layer-wise Recompute (#4526)
* Build Recomputation Graph

* Make topological sort to run FW nodes first

* Pattern match start and end of transformer layer

* Topological sort with Priority

* Add logger to Gradient Graph Builder

* Use Logger

* Introduce Execution Order
2020-09-24 19:56:32 -07:00
..
onnxruntime/core Transformer layer-wise Recompute (#4526) 2020-09-24 19:56:32 -07:00