1 /* Copyright 2020 The TensorFlow Authors. All Rights Reserved. 2 3 Licensed under the Apache License, Version 2.0 (the "License"); 4 you may not use this file except in compliance with the License. 5 You may obtain a copy of the License at 6 7 http://www.apache.org/licenses/LICENSE-2.0 8 9 Unless required by applicable law or agreed to in writing, software 10 distributed under the License is distributed on an "AS IS" BASIS, 11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12 See the License for the specific language governing permissions and 13 limitations under the License. 14 ==============================================================================*/ 15 16 #ifndef TENSORFLOW_LITE_DELEGATES_GPU_COMMON_TRANSFORMATIONS_ADD_QUANT_ADJUSTMENTS_H_ 17 #define TENSORFLOW_LITE_DELEGATES_GPU_COMMON_TRANSFORMATIONS_ADD_QUANT_ADJUSTMENTS_H_ 18 19 #include <memory> 20 21 #include "tensorflow/lite/delegates/gpu/common/model_transformer.h" 22 23 namespace tflite { 24 namespace gpu { 25 26 // This pass is used to support inference on quantized models with the GPU 27 // delegate. 28 // 29 // When delegating quantized models, we still run float-point inference on GPU 30 // under-the-hood. This is done by dequantizing inputs (at runtime) & constants 31 // (during delegation). 32 // However, intermediate tensors can still deviate from the original quantized 33 // inference, since activations may not follow the attributes set by the 34 // original quantizion parameters. 35 // To prevent this, we add "QuantizeAndDequantize" nodes for each node-output 36 // that was originally fixed-point: 37 // op1 -> op2 38 // becomes 39 // op1 -> QuantizeAndDequantize -> op2 40 std::unique_ptr<NodeTransformation> NewAddQuantAdjustments(); 41 42 } // namespace gpu 43 } // namespace tflite 44 45 #endif // TENSORFLOW_LITE_DELEGATES_GPU_COMMON_TRANSFORMATIONS_ADD_QUANT_ADJUSTMENTS_H_ 46