Quantize Callback

Quantize your network during training

---

QuantizeCallback

 QuantizeCallback (qconfig_mapping=None, backend='x86')

Basic class handling tweaks of the training loop by changing a Learner in various events

import timm
pretrained_resnet_34 = timm.create_model('resnet34', pretrained=True)
learn = Learner(dls, pretrained_resnet_34, metrics=accuracy)
learn.model.fc = nn.Linear(512, 2)
learn.fit_one_cycle(10, cbs=QuantizeCallback())

epoch	train_loss	valid_loss	accuracy	time
0	0.593340	0.501450	0.718539	00:12
1	0.395149	0.378658	0.830853	00:11
2	0.247482	0.231792	0.901218	00:11
3	0.164855	0.265116	0.892422	00:11
4	0.112191	0.228632	0.914073	00:11
5	0.070206	0.214058	0.929635	00:11
6	0.050446	0.202638	0.937754	00:11
7	0.033936	0.203362	0.941137	00:11
8	0.024784	0.201417	0.938430	00:11
9	0.022369	0.193307	0.941813	00:12

learn = vision_learner(dls, resnet18, metrics=accuracy)
learn.unfreeze()
learn.fit(5, cbs=QuantizeCallback())

epoch	train_loss	valid_loss	accuracy	time
0	0.571546	0.424050	0.789581	00:06
1	0.469406	0.363851	0.843708	00:06
2	0.407703	0.399239	0.817997	00:06
3	0.375065	0.309377	0.865359	00:06
4	0.323774	0.331475	0.873478	00:06

learn.model

GraphModule(
  (0): Module(
    (0): QuantizedConvReLU2d(3, 64, kernel_size=(7, 7), stride=(2, 2), scale=0.029317768290638924, zero_point=0, padding=(3, 3))
    (3): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
    (4): Module(
      (0): Module(
        (conv1): QuantizedConvReLU2d(64, 64, kernel_size=(3, 3), stride=(1, 1), scale=0.017887497320771217, zero_point=0, padding=(1, 1))
        (conv2): QuantizedConv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), scale=0.0466480627655983, zero_point=66, padding=(1, 1))
      )
      (1): Module(
        (conv1): QuantizedConvReLU2d(64, 64, kernel_size=(3, 3), stride=(1, 1), scale=0.017889995127916336, zero_point=0, padding=(1, 1))
        (conv2): QuantizedConv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), scale=0.07470479607582092, zero_point=66, padding=(1, 1))
      )
    )
    (5): Module(
      (0): Module(
        (conv1): QuantizedConvReLU2d(64, 128, kernel_size=(3, 3), stride=(2, 2), scale=0.0174386166036129, zero_point=0, padding=(1, 1))
        (conv2): QuantizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), scale=0.047718875110149384, zero_point=60, padding=(1, 1))
        (downsample): Module(
          (0): QuantizedConv2d(64, 128, kernel_size=(1, 1), stride=(2, 2), scale=0.04965509846806526, zero_point=68)
        )
      )
      (1): Module(
        (conv1): QuantizedConvReLU2d(128, 128, kernel_size=(3, 3), stride=(1, 1), scale=0.019585009664297104, zero_point=0, padding=(1, 1))
        (conv2): QuantizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), scale=0.05827442184090614, zero_point=70, padding=(1, 1))
      )
    )
    (6): Module(
      (0): Module(
        (conv1): QuantizedConvReLU2d(128, 256, kernel_size=(3, 3), stride=(2, 2), scale=0.02278205193579197, zero_point=0, padding=(1, 1))
        (conv2): QuantizedConv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), scale=0.05654977634549141, zero_point=57, padding=(1, 1))
        (downsample): Module(
          (0): QuantizedConv2d(128, 256, kernel_size=(1, 1), stride=(2, 2), scale=0.019852932542562485, zero_point=75)
        )
      )
      (1): Module(
        (conv1): QuantizedConvReLU2d(256, 256, kernel_size=(3, 3), stride=(1, 1), scale=0.021630365401506424, zero_point=0, padding=(1, 1))
        (conv2): QuantizedConv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), scale=0.06945421546697617, zero_point=73, padding=(1, 1))
      )
    )
    (7): Module(
      (0): Module(
        (conv1): QuantizedConvReLU2d(256, 512, kernel_size=(3, 3), stride=(2, 2), scale=0.019869942218065262, zero_point=0, padding=(1, 1))
        (conv2): QuantizedConv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), scale=0.07700460404157639, zero_point=63, padding=(1, 1))
        (downsample): Module(
          (0): QuantizedConv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), scale=0.045847173780202866, zero_point=68)
        )
      )
      (1): Module(
        (conv1): QuantizedConvReLU2d(512, 512, kernel_size=(3, 3), stride=(1, 1), scale=0.02446889691054821, zero_point=0, padding=(1, 1))
        (conv2): QuantizedConv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), scale=0.3400780260562897, zero_point=54, padding=(1, 1))
      )
    )
  )
  (1): Module(
    (0): Module(
      (mp): AdaptiveMaxPool2d(output_size=1)
      (ap): AdaptiveAvgPool2d(output_size=1)
    )
    (2): BatchNorm1d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (3): QuantizedDropout(p=0.25, inplace=False)
    (4): QuantizedLinearReLU(in_features=1024, out_features=512, scale=0.5987890958786011, zero_point=0, qscheme=torch.per_channel_affine)
    (6): BatchNorm1d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (7): QuantizedDropout(p=0.5, inplace=False)
    (8): QuantizedLinear(in_features=512, out_features=2, scale=0.6221694350242615, zero_point=113, qscheme=torch.per_channel_affine)
  )
)