我试图在AWS GPU(p2-xlarge)上使用ver 1.7.0 tensorflow中的tfdbg。
当我尝试print_tensor或node_info时,我得到一个KeyError:
print_tensor gradients/TopKV2_grad/Gather_1:0
Error occurred during handling of command: print_tensor gradients/TopKV2_grad/Gather_1:0:
<class 'KeyError'>: '/device:CPU:0'
Traceback (most recent call last):
File "/home/usrnm/.local/lib/python3.5/site-packages/tensorflow/python/debug/cli/debugger_cli_common.py", line 666, in dispatch_command
output = handler(argv, screen_info=screen_info)
File "/home/usrnm/.local/lib/python3.5/site-packages/tensorflow/python/debug/cli/analyzer_cli.py", line 930, in print_tensor
watch_keys = self._debug_dump.debug_watch_keys(node_name)
File "/home/usrnm/.local/lib/python3.5/site-packages/tensorflow/python/debug/lib/debug_data.py", line 1365, in debug_watch_keys
if node_name not in self._debug_watches[device_name]:
KeyError: '/device:CPU:0'
我正在运行一个脚本,我没有完成任何设备的手动分配或使用了群集规范,基本上只是让我们处理这些设备。
我试过检查traceback中引用的tf代码但是有点超出我的范围。
为什么tfdbg不满意的任何线索?
谢谢!
编辑#1: 这似乎发生在我
时然后键入