Skip to content

P
Paddle

PaddlePaddle / Paddle
大约 1 年 前同步成功

通知 2298
Star 20931
Fork 5422
P
Paddle
未验证 提交 cbda49e6 编写于 作者: J joanna.wozna.intel 提交者： GitHub
浏览文件
操作

Correct order of passes (#45936)

上级 1711407d
显示空白变更内容
内联 并排
Showing with 3 addition and 2 deletion
python/paddle/fluid/contrib/slim/quantization/quant2_int8_mkldnn_pass.py
浏览文件 @ cbda49e6
...@@ -448,13 +448,14 @@ class Quant2Int8MkldnnPass(object): ...@@ -448,13 +448,14 @@ class Quant2Int8MkldnnPass(object):
graph = self._apply_pass(graph, 'fc_act_mkldnn_fuse_pass') graph = self._apply_pass(graph, 'fc_act_mkldnn_fuse_pass')
graph = self._apply_pass(graph, graph = self._apply_pass(graph,
'matmul_transpose_reshape_mkldnn_fuse_pass') 'matmul_transpose_reshape_mkldnn_fuse_pass')
graph = self._apply_pass(graph,
'matmul_elementwise_add_mkldnn_fuse_pass')
graph = self._apply_pass(graph, 'matmul_activation_mkldnn_fuse_pass')
graph = self._apply_pass(graph, 'batch_norm_act_fuse_pass') graph = self._apply_pass(graph, 'batch_norm_act_fuse_pass')
graph = self._apply_pass(graph, 'softplus_activation_mkldnn_fuse_pass') graph = self._apply_pass(graph, 'softplus_activation_mkldnn_fuse_pass')
graph = self._apply_pass(graph, 'scale_matmul_fuse_pass') graph = self._apply_pass(graph, 'scale_matmul_fuse_pass')
graph = self._apply_pass(graph, graph = self._apply_pass(graph,
'reshape_transpose_matmul_mkldnn_fuse_pass') 'reshape_transpose_matmul_mkldnn_fuse_pass')
graph = self._apply_pass(graph,
'matmul_elementwise_add_mkldnn_fuse_pass')
# the following pass should be the last one since it will work on all fused ops. # the following pass should be the last one since it will work on all fused ops.
graph = self._apply_pass(graph, 'runtime_context_cache_pass') graph = self._apply_pass(graph, 'runtime_context_cache_pass')
return graph return graph
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或