experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Fix comments for the split merged operands pass.

Browse Source 
PiperOrigin-RevId: 339840441
Change-Id: I292db092143c9341026a3b2bae9f1eee7b64e657
This commit is contained in:
Renjie Liu 2020-10-30 03:10:55 -07:00 committed by TensorFlower Gardener
parent 3e12b00d39
commit 5234e72159
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
tensorflow/compiler/mlir/lite/tf_tfl_passes.cc
View File

@ -218,8 +218,8 @@ void AddTFToTFLConversionPasses(const mlir::TFL::PassConfig& pass_config,
AddQuantizationPasses(pass_config.quant_specs, pass_manager);
}
// This pass should be always at the end of the floating point model
// conversion. Some TFL ops like unidirectional
// This pass should be always at the end of the model
// conversion (even after quantization). Some TFL ops like unidirectional
// sequence lstm will have stateful operands and some optimization passes
// will merge those operands if they have identical values & types. However,
// it's not desired by TFL. This pass serves as a "fix" pass to split the