10 Commits (2de5075a6d669640bdcd9525cfc3eab0c8595c6e)

Author	SHA1	Message	Date
Liufang Sang	915b892a15	Fix div zero in fake quantize op (#22966) ... * fix div zero test=develop * fix div zero test=develop * add hostdevice function test=develop * add eps when is zero test=develop	5 years ago
Huihuang Zheng	12542320c5	Replace TemporaryAllocator by CUDADeviceContextAllocator (#18989) ... TemporaryAllocator is a singleton used for allocating memory for Cudnn. Since it is a singleton, we can delete it for better performance in memory. We replace TemporaryAllocator by CUDADeviceContextAllocator and CUDADeviceContextAllocation, which uses stream callback to delete the memory allocated for the stream to avoid singleton. Also added data_feed_proto to operator to fix CI in CPU compilation	6 years ago
Zhaolong Xing	ff7f911b4d	add quant_dequant_moving_avg_max_abs op (#17480) ... * add quant_dequant_moving_avg_max_abs op test=develop * add more note for quantdequant op test=develop	6 years ago
Zhen Wang	a914d9b116	Quant output scale (#17215) ... * Add MovingAverageAbsMaxScale operator which is only used for calculating the quantization scale. * test=develop * change the output into inplace. test=develop * Revert "test=develop" This reverts commit 696cf62699ba1e1c98f61f7345ac7060010eb29a. * Revert "change the output into inplace. test=develop" This reverts commit a19acd20f07eee82622701a3015e6e9c073a5e0b. * test=develop. * update the MovingAverageAbsMaxScaleOp test. test=develop	6 years ago
Zhen Wang	8965819fbb	rewrite the cuda kernels of channel_wise_quant_op and channe_wise_dequant_op. test=develop	6 years ago
Zhen Wang	ec88b6cc5a	add channel wise quantization in ir pass.	6 years ago
achao2013	81b4fad8b9	add moving average absmax op and fix bug (#15155) ... * Add moving average absmax op in quantilize-aware training.	6 years ago
Zhen Wang	545247d7b4	add channel wise quantize op.	6 years ago
qingqing01	9bd933d3fb	Improve and fix fake_quantize_op (#13092) ... * Improve and fix fake_quantize_op.	7 years ago
achao2013	8e4b225fe4	Add fake_quantize_op. (#11359) ... * Add a fake_quantize_op, which quantize an input tensor to a tensor with lower bits.	7 years ago