名称 最后提交 最后更新
..
winograd Fix undefined symbol for ios arm v8
activation.h Refactor dequant fusion kernels to support more fusion patterns
activation_functions.h merge nlp to main
conv_func.h fix docker complile error
depthwise_conv3x3.cpp Support padding in 8bit depthwise conv, so remove padding from dequantize kernel
depthwise_conv3x3.h Optimize int8 depthwise conv
depthwise_conv3x3_int8.cpp Optimize int8 depthwise conv
depthwise_conv3x3_int8_arm64.cpp Fix undefined symbol for ios arm v8
elementwise_op_function.h code style
gemm.cpp Optimize: fuse quantize and pad op
gemm.h add fusion fc int8_t op and its UT.
gemm_int8.cpp add fusion fc int8_t op and its UT.
gemm_omp_int8.cpp add int8_t type sgemm_omp
gpc.cpp add multi-point NMS
gpc.h add multi-point NMS
gru_compute.cpp fix some bugs.
gru_compute.h merge nlp to main
gru_cpu_kernel.h merge nlp to main
gru_kernel.h merge nlp to main
im2col.cpp Resolve merge conflicts
im2col.h add copyright
math_func_neon.h fix #224
math_function.cpp Change 'val * (1.f / count)' to 'val / count' to fix average pooling calculation precision
math_function.h Change 'val * (1.f / count)' to 'val / count' to fix average pooling calculation precision
math_function_int8.cpp Change 'val * (1.f / count)' to 'val / count' to fix average pooling calculation precision
pad.cpp
pad.h
poly_util.cpp
poly_util.h
pooling.cpp
pooling.h
pooling3x3.cpp
quantize.h
selected_rows_functor.h
sequence2batch.cpp
sequence2batch.h
softmax.cpp
softmax.h
transform.h
vol2col.cpp
vol2col.h

项目简介

Multi-platform high performance deep learning inference engine (『飞桨』多平台高性能深度学习预测引擎)

发行版本 20

v2.7-beta

全部发行版

贡献者 87

全部贡献者

开发语言

  • C++ 82.3 %
  • Swift 4.1 %
  • CMake 3.0 %
  • Metal 2.6 %
  • C 2.3 %