未验证 提交 27b42183 编写于 作者: H Huihuang Zheng 提交者: GitHub

[Dy2stat] Disable PaddleInference IR Optimization in test_mnist for CUDA11 (#29105)

test_mnist failed on CUDA11. We found that it is due to PaddleInference IR Optimization after debugging. We disable it in this PR and we will re-enable it after PaddleInference fixes it.
上级 01bdea7c
file(GLOB TEST_OPS RELATIVE "${CMAKE_CURRENT_SOURCE_DIR}" "test_*.py") file(GLOB TEST_OPS RELATIVE "${CMAKE_CURRENT_SOURCE_DIR}" "test_*.py")
string(REPLACE ".py" "" TEST_OPS "${TEST_OPS}") string(REPLACE ".py" "" TEST_OPS "${TEST_OPS}")
# disable for cuda11
list(REMOVE_ITEM TEST_OPS test_mnist)
list(REMOVE_ITEM TEST_OPS test_resnet)
foreach(TEST_OP ${TEST_OPS}) foreach(TEST_OP ${TEST_OPS})
py_test_modules(${TEST_OP} MODULES ${TEST_OP}) py_test_modules(${TEST_OP} MODULES ${TEST_OP})
endforeach(TEST_OP) endforeach(TEST_OP)
......
...@@ -57,7 +57,9 @@ class PredictorTools(object): ...@@ -57,7 +57,9 @@ class PredictorTools(object):
config.switch_use_feed_fetch_ops(False) config.switch_use_feed_fetch_ops(False)
config.enable_memory_optim() config.enable_memory_optim()
config.disable_glog_info() config.disable_glog_info()
config.switch_ir_optim(True) # TODO: set it to True after PaddleInference fix the precision error
# in CUDA11
config.switch_ir_optim(False)
return config return config
......
Markdown is supported
0% .
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