Size mismatch error

[winstondcosta]

Traceback (most recent call last):
File "msunas.py", line 355, in
main(cfgs)
File "msunas.py", line 310, in main
engine.search()
File "msunas.py", line 83, in search
candidates, c_top1_err_pred = self._next(archive, acc_predictor, self.n_iter)
File "msunas.py", line 197, in _next
{'n_classes': self.n_classes, 'model_path': self.supernet_path})
File "msunas.py", line 266, in init
n_classes=supernet['n_classes'], model_path=supernet['model_path'])
File "C:\Users\hp\nsganetv2-master\evaluator.py", line 102, in init
self.engine.load_weights_from_net(init)
File "E:\Installed_Softwares\lib\site-packages\ofa\elastic_nn\networks\ofa_mbv3.py", line 211, in load_weights_from_net
self.load_state_dict(model_dict)
File "E:\Installed_Softwares\lib\site-packages\torch\nn\modules\module.py", line 1052, in load_state_dict
self.class.name, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for OFAMobileNetV3:
size mismatch for classifier.linear.weight: copying a param with shape torch.Size([1000, 1280]) from checkpoint, the shape in current model is torch.Size([10, 1280]).
size mismatch for classifier.linear.bias: copying a param with shape torch.Size([1000]) from checkpoint, the shape in current model is torch.Size([10]).

[alex-stoica]

I had the same issue, I suppose you tried the bi-objective search code with the argument --n_classes=10.
A quick ugly fix which worked for me was changing some lines of code from evaluator.py, after initialization of init variable:
init = torch.load(model_path, map_location='cpu')['state_dict']
Init is a dict having a classifier with 1000 outputs:
classifier.linear.weight torch.Size([1000, 1280])
classifier.linear.bias torch.Size([1000])
Here is the place you may do some changes. Just for illustration purposes, modifying

init['classifier.linear.weight'] = init['classifier.linear.weight'][:10]
init['classifier.linear.bias'] = init['classifier.linear.bias'][:10]

resolves your problem.

[euminds]

Did you encounter this kind of error again when searching? An error is reported when the configured ofa version is ofa 0.1.0-202012082159. Then try ofa 0.0.4-2012082155 but the same error still occurs.
Traceback (most recent call last):
File "msunas.py", line 8, in
from evaluator import OFAEvaluator, get_net_info
File "/data8T/nsganetv2-master/evaluator.py", line 8, in
from codebase.networks import NSGANetV2
File "/data8T/nsganetv2-master/codebase/networks/init.py", line 1, in
from ofa.imagenet_codebase.networks.proxyless_nets import ProxylessNASNets, proxyless_base, MobileNetV2
ModuleNotFoundError: No module named 'ofa.imagenet_codebase'

[alex-stoica]

@euminds - I had plenty of imports errors related with ofa.
I solved them by looking at their github repo: https://github.com/mit-han-lab/once-for-all
If you have the last version of ofa, it's easy to look to your required files.
In your specific case,
from ofa.imagenet_codebase.networks.proxyless_nets import ProxylessNASNets, proxyless_base, MobileNetV2
shoud be replaced with
from ofa.imagenet_classification.networks.proxyless_nets import ProxylessNASNets, proxyless_base, MobileNetV2
If you check the github repo, you see that there is no folder "imagenet_codebase", but there is a folder "imagenet_classification", which contains what you need

[euminds]

@euminds - I had plenty of imports errors related with ofa.
I solved them by looking at their github repo: https://github.com/mit-han-lab/once-for-all
If you have the last version of ofa, it's easy to look to your required files.
In your specific case,
from ofa.imagenet_codebase.networks.proxyless_nets import ProxylessNASNets, proxyless_base, MobileNetV2
shoud be replaced with
from ofa.imagenet_classification.networks.proxyless_nets import ProxylessNASNets, proxyless_base, MobileNetV2
If you check the github repo, you see that there is no folder "imagenet_codebase", but there is a folder "imagenet_classification", which contains what you need

Thanks. I will try.