Update release doc #9729
Update release doc #9729
Conversation
| <img src="https://raw.githubusercontent.com/PaddlePaddle/Paddle/develop/doc/fluid/images/ci_build_whl.png"> | ||
| 弹出下面的选择框，在第二个tab (Changes)里选择需要发布的分支，这里选择0.11.0，然后点击"Run Build"按钮。 | ||
| <img src="https://raw.githubusercontent.com/PaddlePaddle/Paddle/develop/doc/fluid/images/ci_build_whl.png"> | ||
| 1. 等待编译完成后可以在此页面的"Artifacts"下拉框中找到生成的3个二进制文件，分别对应CAPI，`cp27m`和`cp27mu`的版本。 |
There was a problem hiding this comment.
不确定，看后续是否有capi的计划，如果都是TensorRT可能会不需要。
There was a problem hiding this comment.
Several approaches in my mind for Fluid to support inference:
-
The Fluid inference library created by @Xreki @kexinzhao @sidgoyal78 @luotao1 , is the leading effort of other approaches. This approach creates an inference library that can be built for x64 and ARM.
-
The MKLDNN-based kernels contributed by Intel, with contact person @luotao1 , that would optimize approach 1.'s performance with Intel Xeon CPU.
-
The capability for Fluid to export an ONNX-format model, which could pipe with TensorRT 4.x to generate CUDA-optimized inference library.
-
Other approaches, including bridging ProgramDesc to Intel's ngraph, and MDL, an ARM-based inference library for Fluid that loads and executes a ProgramDesc message.
3 participants
Resolves #9573