SqueezeNet 1.0 and 1.1 models

[Maratyszcza]

SqueezeNet 1.0 and 1.1 models. I'm still working on getting them converge on ImageNet, but as several people contacted me about SqueezeNet for PyTorch, I think its useful to share it publicly.

[Maratyszcza]

I successfully trained SqueezeNet 1.0 on ImageNet with the following parameters: --arch squeezenet1_0 --epochs 90 --lr 0.01 --weight-decay 0.0002 --batch-size 128.
Accuracy: Prec@1 57.868 Prec@5 80.424

[apaszke]

Awesome! Is the PR ready?

[Maratyszcza]

There is SqueezeNet 1.1 to train as well

[Maratyszcza]

Finished training both SqueezeNet 1.0 and 1.1. Trained for 100 epochs with --epochs 100 --lr 0.01 --weight-decay 0.0002 --batch-size 128

Model	Top-1 accuracy	Top-5 accuracy
SqueezeNet 1.0	58.000%	80.488%
SqueezeNet 1.1	58.184%	80.514%

After @soumith uploads the models to AWS, I'll add model_zoo links, and we'd be ready to merge.

[apaszke]

Great! If you put them somewhere or send them to me, I can upload them to AWS too.

[soumith]

finished uploading them:

https://s3.amazonaws.com/pytorch/models/squeezenet1_0-0ef9bd12.pth.tar
https://s3.amazonaws.com/pytorch/models/squeezenet1_1-f0e1d24d.pth.tar

[Maratyszcza]

@apaszke @soumith Ready to merge!

[apaszke]

Two minor things. Ready to merge after fixing them.