With the current torch version, torch._six does not exist anymore.
A workaround would be to directly set, in utils/my_data_loader.py:
replace from torch._six import string_classes by string_classes = str
Loading the LatencyTable also seems to generate an error due to an updated yaml package.
I modified: yaml.load(fp) to yaml.full_load(fp) in latency_table.py
I am also curious if you're interested in including more hardware platforms in your list. I trained and searched OFA results on Raspberry Pi and Redmi Note 7S.
Good evening!
With the current torch version,
torch._sixdoes not exist anymore. A workaround would be to directly set, in utils/my_data_loader.py: replacefrom torch._six import string_classesbystring_classes = strLoading the LatencyTable also seems to generate an error due to an updated yaml package. I modified:
yaml.load(fp)toyaml.full_load(fp)in latency_table.pyI am also curious if you're interested in including more hardware platforms in your list. I trained and searched OFA results on Raspberry Pi and Redmi Note 7S.
Best,