iamkanghyunchoi / ait Goto Github PK

Thanks for your release of the code.

When I quantize resnet18 to 4bit with Qimera and AIT and then evaluate the model on imagenet, an abnormal intermediate result is obtained:

2022-09-13 15:14:24,225 INFO: #==>Best Result of ep 60 is: Top1 Accuracy: 0.100000, Top5 Accuracy: 0.500000 at ep 60
2022-09-13 15:20:09,837 INFO: #==>[Epoch 61/400] [acc: 0.093750] [train loss: nan]
2022-09-13 15:20:09,838 INFO: #==>Best Result of ep 61 is: Top1 Accuracy: 0.100000, Top5 Accuracy: 0.500000 at ep 61
2022-09-13 15:25:58,151 INFO: #==>[Epoch 62/400] [acc: 0.125000] [train loss: nan]
2022-09-13 15:25:58,153 INFO: #==>Best Result of ep 62 is: Top1 Accuracy: 0.100000, Top5 Accuracy: 0.500000 at ep 62
2022-09-13 15:31:30,271 INFO: #==>[Epoch 63/400] [acc: 0.062500] [train loss: nan]
2022-09-13 15:31:30,273 INFO: #==>Best Result of ep 63 is: Top1 Accuracy: 0.100000, Top5 Accuracy: 0.500000 at ep 63
2022-09-13 15:37:26,569 INFO: #==>[Epoch 64/400] [acc: 0.031250] [train loss: nan]

It seems that the model training does not converge.

My name is Mingi Yoo.
I saw your code.
Your code is pretty good.
I want to pick you up.
Have a nice day:)

When i played with ZeroQ codes. I was very surprised to find that the results for ZeroQ/Resnet-18/w4a4/ in Table 1 of the paper are very low, 22.58%. But when I tried to run the official ZeroQ code, the result was 47.96%. Result for w5a5 setting is 68.26 % compared to 59.6 in the paper.

By the way, your work is solid and impressive!!!

Thanks for your good work on data-free quantization and the release of the code.

Zero-shot adversarial quantization(ZAQ) is also a data-free quantization method with adversarial exploration. I notice that both IntraQ and AIT are not compared to the ZAQ method. Are they solving different problems or having different experiment settings?

Thanks for your answer.

Python >= 3.7 has dataclasses as a built-in module. So, there is a conflict while executing pip install -r requirements.txt cmd.

Both numpy and typing-extension are outdated version, which leads to make some conflicts with recently released PyTorch or other ML frameworks.

For my case with torch1.10.1+cu111, numpy==1.23.1 and typing-extensions=4.1.1 are stable.