modelId
stringlengths 4
112
| lastModified
stringlengths 24
24
| tags
list | pipeline_tag
stringclasses 21
values | files
list | publishedBy
stringlengths 2
37
| downloads_last_month
int32 0
9.44M
| library
stringclasses 15
values | modelCard
large_stringlengths 0
100k
|
|---|---|---|---|---|---|---|---|---|
glasses/dummy | 2021-04-21T18:24:15.000Z | [
"pytorch",
"arxiv:1512.03385",
"arxiv:1812.01187",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 17 | transformers | # ResNet
Implementation of ResNet proposed in [Deep Residual Learning for Image
Recognition](https://arxiv.org/abs/1512.03385)
``` python
ResNet.resnet18()
ResNet.resnet26()
ResNet.resnet34()
ResNet.resnet50()
ResNet.resnet101()
ResNet.resnet152()
ResNet.resnet200()
Variants (d) proposed in `Bag of Tricks for Image Classification with Convolutional Neural Networks <https://arxiv.org/pdf/1812.01187.pdf`_
ResNet.resnet26d()
ResNet.resnet34d()
ResNet.resnet50d()
# You can construct your own one by chaning `stem` and `block`
resnet101d = ResNet.resnet101(stem=ResNetStemC, block=partial(ResNetBottleneckBlock, shortcut=ResNetShorcutD))
```
Examples:
``` python
# change activation
ResNet.resnet18(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNet.resnet18(n_classes=100)
# pass a different block
ResNet.resnet18(block=SENetBasicBlock)
# change the steam
model = ResNet.resnet18(stem=ResNetStemC)
change shortcut
model = ResNet.resnet18(block=partial(ResNetBasicBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = ResNet.resnet18()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
|
glasses/eca_resnet101d | 2021-04-25T07:55:15.000Z | [
"pytorch",
"transformers"
]
| [
".gitattributes",
"config.json",
"pytorch_model.bin"
]
| glasses | 7 | transformers | ||
glasses/eca_resnet26t | 2021-04-24T10:53:05.000Z | [
"pytorch",
"transformers"
]
| [
".gitattributes",
"config.json",
"pytorch_model.bin"
]
| glasses | 10 | transformers | ||
glasses/eca_resnet50d | 2021-04-25T07:54:33.000Z | [
"pytorch",
"transformers"
]
| [
".gitattributes",
"config.json",
"pytorch_model.bin"
]
| glasses | 7 | transformers | ||
glasses/eca_resnet50t | 2021-04-24T10:53:17.000Z | [
"pytorch",
"transformers"
]
| [
".gitattributes",
"config.json",
"pytorch_model.bin"
]
| glasses | 7 | transformers | ||
glasses/efficientnet_b0 | 2021-04-22T17:59:42.000Z | [
"pytorch",
"arxiv:1905.11946",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 13 | transformers | # efficientnet_b0
Implementation of EfficientNet proposed in [EfficientNet: Rethinking
Model Scaling for Convolutional Neural
Networks](https://arxiv.org/abs/1905.11946)
The basic architecture is similar to MobileNetV2 as was computed by
using [Progressive Neural Architecture
Search](https://arxiv.org/abs/1905.11946) .
The following table shows the basic architecture
(EfficientNet-efficientnet\_b0):
Then, the architecture is scaled up from
[-efficientnet\_b0]{.title-ref} to [-efficientnet\_b7]{.title-ref}
using compound scaling.
``` python
EfficientNet.efficientnet_b0()
EfficientNet.efficientnet_b1()
EfficientNet.efficientnet_b2()
EfficientNet.efficientnet_b3()
EfficientNet.efficientnet_b4()
EfficientNet.efficientnet_b5()
EfficientNet.efficientnet_b6()
EfficientNet.efficientnet_b7()
EfficientNet.efficientnet_b8()
EfficientNet.efficientnet_l2()
```
Examples:
``` python
EfficientNet.efficientnet_b0(activation = nn.SELU)
# change number of classes (default is 1000 )
EfficientNet.efficientnet_b0(n_classes=100)
# pass a different block
EfficientNet.efficientnet_b0(block=...)
# store each feature
x = torch.rand((1, 3, 224, 224))
model = EfficientNet.efficientnet_b0()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
# [torch.Size([1, 32, 112, 112]), torch.Size([1, 24, 56, 56]), torch.Size([1, 40, 28, 28]), torch.Size([1, 80, 14, 14])]
```
|
|
glasses/efficientnet_b1 | 2021-04-22T17:59:47.000Z | [
"pytorch",
"arxiv:1905.11946",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 7 | transformers | # efficientnet_b1
Implementation of EfficientNet proposed in [EfficientNet: Rethinking
Model Scaling for Convolutional Neural
Networks](https://arxiv.org/abs/1905.11946)
The basic architecture is similar to MobileNetV2 as was computed by
using [Progressive Neural Architecture
Search](https://arxiv.org/abs/1905.11946) .
The following table shows the basic architecture
(EfficientNet-efficientnet\_b0):
Then, the architecture is scaled up from
[-efficientnet\_b0]{.title-ref} to [-efficientnet\_b7]{.title-ref}
using compound scaling.
``` python
EfficientNet.efficientnet_b0()
EfficientNet.efficientnet_b1()
EfficientNet.efficientnet_b2()
EfficientNet.efficientnet_b3()
EfficientNet.efficientnet_b4()
EfficientNet.efficientnet_b5()
EfficientNet.efficientnet_b6()
EfficientNet.efficientnet_b7()
EfficientNet.efficientnet_b8()
EfficientNet.efficientnet_l2()
```
Examples:
``` python
EfficientNet.efficientnet_b0(activation = nn.SELU)
# change number of classes (default is 1000 )
EfficientNet.efficientnet_b0(n_classes=100)
# pass a different block
EfficientNet.efficientnet_b0(block=...)
# store each feature
x = torch.rand((1, 3, 224, 224))
model = EfficientNet.efficientnet_b0()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
# [torch.Size([1, 32, 112, 112]), torch.Size([1, 24, 56, 56]), torch.Size([1, 40, 28, 28]), torch.Size([1, 80, 14, 14])]
```
|
|
glasses/efficientnet_b2 | 2021-04-22T17:59:53.000Z | [
"pytorch",
"arxiv:1905.11946",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 6 | transformers | # efficientnet_b2
Implementation of EfficientNet proposed in [EfficientNet: Rethinking
Model Scaling for Convolutional Neural
Networks](https://arxiv.org/abs/1905.11946)
The basic architecture is similar to MobileNetV2 as was computed by
using [Progressive Neural Architecture
Search](https://arxiv.org/abs/1905.11946) .
The following table shows the basic architecture
(EfficientNet-efficientnet\_b0):
Then, the architecture is scaled up from
[-efficientnet\_b0]{.title-ref} to [-efficientnet\_b7]{.title-ref}
using compound scaling.
``` python
EfficientNet.efficientnet_b0()
EfficientNet.efficientnet_b1()
EfficientNet.efficientnet_b2()
EfficientNet.efficientnet_b3()
EfficientNet.efficientnet_b4()
EfficientNet.efficientnet_b5()
EfficientNet.efficientnet_b6()
EfficientNet.efficientnet_b7()
EfficientNet.efficientnet_b8()
EfficientNet.efficientnet_l2()
```
Examples:
``` python
EfficientNet.efficientnet_b0(activation = nn.SELU)
# change number of classes (default is 1000 )
EfficientNet.efficientnet_b0(n_classes=100)
# pass a different block
EfficientNet.efficientnet_b0(block=...)
# store each feature
x = torch.rand((1, 3, 224, 224))
model = EfficientNet.efficientnet_b0()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
# [torch.Size([1, 32, 112, 112]), torch.Size([1, 24, 56, 56]), torch.Size([1, 40, 28, 28]), torch.Size([1, 80, 14, 14])]
```
|
|
glasses/efficientnet_b3 | 2021-04-22T17:59:59.000Z | [
"pytorch",
"arxiv:1905.11946",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 48 | transformers | # efficientnet_b3
Implementation of EfficientNet proposed in [EfficientNet: Rethinking
Model Scaling for Convolutional Neural
Networks](https://arxiv.org/abs/1905.11946)
The basic architecture is similar to MobileNetV2 as was computed by
using [Progressive Neural Architecture
Search](https://arxiv.org/abs/1905.11946) .
The following table shows the basic architecture
(EfficientNet-efficientnet\_b0):
Then, the architecture is scaled up from
[-efficientnet\_b0]{.title-ref} to [-efficientnet\_b7]{.title-ref}
using compound scaling.
``` python
EfficientNet.efficientnet_b0()
EfficientNet.efficientnet_b1()
EfficientNet.efficientnet_b2()
EfficientNet.efficientnet_b3()
EfficientNet.efficientnet_b4()
EfficientNet.efficientnet_b5()
EfficientNet.efficientnet_b6()
EfficientNet.efficientnet_b7()
EfficientNet.efficientnet_b8()
EfficientNet.efficientnet_l2()
```
Examples:
``` python
EfficientNet.efficientnet_b0(activation = nn.SELU)
# change number of classes (default is 1000 )
EfficientNet.efficientnet_b0(n_classes=100)
# pass a different block
EfficientNet.efficientnet_b0(block=...)
# store each feature
x = torch.rand((1, 3, 224, 224))
model = EfficientNet.efficientnet_b0()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
# [torch.Size([1, 32, 112, 112]), torch.Size([1, 24, 56, 56]), torch.Size([1, 40, 28, 28]), torch.Size([1, 80, 14, 14])]
```
|
|
glasses/efficientnet_b5 | 2021-04-22T18:00:08.000Z | [
"pytorch",
"arxiv:1905.11946",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 11 | transformers | # efficientnet_b5
Implementation of EfficientNet proposed in [EfficientNet: Rethinking
Model Scaling for Convolutional Neural
Networks](https://arxiv.org/abs/1905.11946)
The basic architecture is similar to MobileNetV2 as was computed by
using [Progressive Neural Architecture
Search](https://arxiv.org/abs/1905.11946) .
The following table shows the basic architecture
(EfficientNet-efficientnet\_b0):
Then, the architecture is scaled up from
[-efficientnet\_b0]{.title-ref} to [-efficientnet\_b7]{.title-ref}
using compound scaling.
``` python
EfficientNet.efficientnet_b0()
EfficientNet.efficientnet_b1()
EfficientNet.efficientnet_b2()
EfficientNet.efficientnet_b3()
EfficientNet.efficientnet_b4()
EfficientNet.efficientnet_b5()
EfficientNet.efficientnet_b6()
EfficientNet.efficientnet_b7()
EfficientNet.efficientnet_b8()
EfficientNet.efficientnet_l2()
```
Examples:
``` python
EfficientNet.efficientnet_b0(activation = nn.SELU)
# change number of classes (default is 1000 )
EfficientNet.efficientnet_b0(n_classes=100)
# pass a different block
EfficientNet.efficientnet_b0(block=...)
# store each feature
x = torch.rand((1, 3, 224, 224))
model = EfficientNet.efficientnet_b0()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
# [torch.Size([1, 32, 112, 112]), torch.Size([1, 24, 56, 56]), torch.Size([1, 40, 28, 28]), torch.Size([1, 80, 14, 14])]
```
|
|
glasses/efficientnet_b6 | 2021-04-22T18:00:19.000Z | [
"pytorch",
"arxiv:1905.11946",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 7 | transformers | # efficientnet_b6
Implementation of EfficientNet proposed in [EfficientNet: Rethinking
Model Scaling for Convolutional Neural
Networks](https://arxiv.org/abs/1905.11946)
The basic architecture is similar to MobileNetV2 as was computed by
using [Progressive Neural Architecture
Search](https://arxiv.org/abs/1905.11946) .
The following table shows the basic architecture
(EfficientNet-efficientnet\_b0):
Then, the architecture is scaled up from
[-efficientnet\_b0]{.title-ref} to [-efficientnet\_b7]{.title-ref}
using compound scaling.
``` python
EfficientNet.efficientnet_b0()
EfficientNet.efficientnet_b1()
EfficientNet.efficientnet_b2()
EfficientNet.efficientnet_b3()
EfficientNet.efficientnet_b4()
EfficientNet.efficientnet_b5()
EfficientNet.efficientnet_b6()
EfficientNet.efficientnet_b7()
EfficientNet.efficientnet_b8()
EfficientNet.efficientnet_l2()
```
Examples:
``` python
EfficientNet.efficientnet_b0(activation = nn.SELU)
# change number of classes (default is 1000 )
EfficientNet.efficientnet_b0(n_classes=100)
# pass a different block
EfficientNet.efficientnet_b0(block=...)
# store each feature
x = torch.rand((1, 3, 224, 224))
model = EfficientNet.efficientnet_b0()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
# [torch.Size([1, 32, 112, 112]), torch.Size([1, 24, 56, 56]), torch.Size([1, 40, 28, 28]), torch.Size([1, 80, 14, 14])]
```
|
|
glasses/efficientnet_lite0 | 2021-04-25T13:49:00.000Z | [
"pytorch",
"transformers"
]
| [
".gitattributes",
"config.json",
"pytorch_model.bin"
]
| glasses | 12 | transformers | ||
glasses/regnetx_002 | 2021-04-21T19:07:29.000Z | [
"pytorch",
"arxiv:2003.13678",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 8 | transformers | # regnetx_002
Implementation of RegNet proposed in [Designing Network Design
Spaces](https://arxiv.org/abs/2003.13678)
The main idea is to start with a high dimensional search space and
iteratively reduce the search space by empirically apply constrains
based on the best performing models sampled by the current search
space.
The resulting models are light, accurate, and faster than
EfficientNets (up to 5x times!)
For example, to go from $AnyNet_A$ to $AnyNet_B$ they fixed the
bottleneck ratio $b_i$ for all stage $i$. The following table shows
all the restrictions applied from one search space to the next one.
The paper is really well written and very interesting, I highly
recommended read it.
``` python
ResNet.regnetx_002()
ResNet.regnetx_004()
ResNet.regnetx_006()
ResNet.regnetx_008()
ResNet.regnetx_016()
ResNet.regnetx_040()
ResNet.regnetx_064()
ResNet.regnetx_080()
ResNet.regnetx_120()
ResNet.regnetx_160()
ResNet.regnetx_320()
# Y variants (with SE)
ResNet.regnety_002()
# ...
ResNet.regnetx_320()
You can easily customize your model
```
Examples:
``` python
# change activation
RegNet.regnetx_004(activation = nn.SELU)
# change number of classes (default is 1000 )
RegNet.regnetx_004(n_classes=100)
# pass a different block
RegNet.regnetx_004(block=RegNetYBotteneckBlock)
# change the steam
model = RegNet.regnetx_004(stem=ResNetStemC)
change shortcut
model = RegNet.regnetx_004(block=partial(RegNetYBotteneckBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = RegNet.regnetx_004()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 32, 112, 112]), torch.Size([1, 32, 56, 56]), torch.Size([1, 64, 28, 28]), torch.Size([1, 160, 14, 14])]
```
|
|
glasses/regnetx_004 | 2021-04-21T19:07:34.000Z | [
"pytorch",
"arxiv:2003.13678",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 11 | transformers | # regnetx_004
Implementation of RegNet proposed in [Designing Network Design
Spaces](https://arxiv.org/abs/2003.13678)
The main idea is to start with a high dimensional search space and
iteratively reduce the search space by empirically apply constrains
based on the best performing models sampled by the current search
space.
The resulting models are light, accurate, and faster than
EfficientNets (up to 5x times!)
For example, to go from $AnyNet_A$ to $AnyNet_B$ they fixed the
bottleneck ratio $b_i$ for all stage $i$. The following table shows
all the restrictions applied from one search space to the next one.
The paper is really well written and very interesting, I highly
recommended read it.
``` python
ResNet.regnetx_002()
ResNet.regnetx_004()
ResNet.regnetx_006()
ResNet.regnetx_008()
ResNet.regnetx_016()
ResNet.regnetx_040()
ResNet.regnetx_064()
ResNet.regnetx_080()
ResNet.regnetx_120()
ResNet.regnetx_160()
ResNet.regnetx_320()
# Y variants (with SE)
ResNet.regnety_002()
# ...
ResNet.regnetx_320()
You can easily customize your model
```
Examples:
``` python
# change activation
RegNet.regnetx_004(activation = nn.SELU)
# change number of classes (default is 1000 )
RegNet.regnetx_004(n_classes=100)
# pass a different block
RegNet.regnetx_004(block=RegNetYBotteneckBlock)
# change the steam
model = RegNet.regnetx_004(stem=ResNetStemC)
change shortcut
model = RegNet.regnetx_004(block=partial(RegNetYBotteneckBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = RegNet.regnetx_004()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 32, 112, 112]), torch.Size([1, 32, 56, 56]), torch.Size([1, 64, 28, 28]), torch.Size([1, 160, 14, 14])]
```
|
|
glasses/regnetx_006 | 2021-04-21T19:07:40.000Z | [
"pytorch",
"arxiv:2003.13678",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 10 | transformers | # regnetx_006
Implementation of RegNet proposed in [Designing Network Design
Spaces](https://arxiv.org/abs/2003.13678)
The main idea is to start with a high dimensional search space and
iteratively reduce the search space by empirically apply constrains
based on the best performing models sampled by the current search
space.
The resulting models are light, accurate, and faster than
EfficientNets (up to 5x times!)
For example, to go from $AnyNet_A$ to $AnyNet_B$ they fixed the
bottleneck ratio $b_i$ for all stage $i$. The following table shows
all the restrictions applied from one search space to the next one.
The paper is really well written and very interesting, I highly
recommended read it.
``` python
ResNet.regnetx_002()
ResNet.regnetx_004()
ResNet.regnetx_006()
ResNet.regnetx_008()
ResNet.regnetx_016()
ResNet.regnetx_040()
ResNet.regnetx_064()
ResNet.regnetx_080()
ResNet.regnetx_120()
ResNet.regnetx_160()
ResNet.regnetx_320()
# Y variants (with SE)
ResNet.regnety_002()
# ...
ResNet.regnetx_320()
You can easily customize your model
```
Examples:
``` python
# change activation
RegNet.regnetx_004(activation = nn.SELU)
# change number of classes (default is 1000 )
RegNet.regnetx_004(n_classes=100)
# pass a different block
RegNet.regnetx_004(block=RegNetYBotteneckBlock)
# change the steam
model = RegNet.regnetx_004(stem=ResNetStemC)
change shortcut
model = RegNet.regnetx_004(block=partial(RegNetYBotteneckBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = RegNet.regnetx_004()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 32, 112, 112]), torch.Size([1, 32, 56, 56]), torch.Size([1, 64, 28, 28]), torch.Size([1, 160, 14, 14])]
```
|
|
glasses/regnetx_008 | 2021-04-21T19:07:45.000Z | [
"pytorch",
"arxiv:2003.13678",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 9 | transformers | # regnetx_008
Implementation of RegNet proposed in [Designing Network Design
Spaces](https://arxiv.org/abs/2003.13678)
The main idea is to start with a high dimensional search space and
iteratively reduce the search space by empirically apply constrains
based on the best performing models sampled by the current search
space.
The resulting models are light, accurate, and faster than
EfficientNets (up to 5x times!)
For example, to go from $AnyNet_A$ to $AnyNet_B$ they fixed the
bottleneck ratio $b_i$ for all stage $i$. The following table shows
all the restrictions applied from one search space to the next one.
The paper is really well written and very interesting, I highly
recommended read it.
``` python
ResNet.regnetx_002()
ResNet.regnetx_004()
ResNet.regnetx_006()
ResNet.regnetx_008()
ResNet.regnetx_016()
ResNet.regnetx_040()
ResNet.regnetx_064()
ResNet.regnetx_080()
ResNet.regnetx_120()
ResNet.regnetx_160()
ResNet.regnetx_320()
# Y variants (with SE)
ResNet.regnety_002()
# ...
ResNet.regnetx_320()
You can easily customize your model
```
Examples:
``` python
# change activation
RegNet.regnetx_004(activation = nn.SELU)
# change number of classes (default is 1000 )
RegNet.regnetx_004(n_classes=100)
# pass a different block
RegNet.regnetx_004(block=RegNetYBotteneckBlock)
# change the steam
model = RegNet.regnetx_004(stem=ResNetStemC)
change shortcut
model = RegNet.regnetx_004(block=partial(RegNetYBotteneckBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = RegNet.regnetx_004()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 32, 112, 112]), torch.Size([1, 32, 56, 56]), torch.Size([1, 64, 28, 28]), torch.Size([1, 160, 14, 14])]
```
|
|
glasses/regnetx_016 | 2021-04-21T19:07:51.000Z | [
"pytorch",
"arxiv:2003.13678",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 9 | transformers | # regnetx_016
Implementation of RegNet proposed in [Designing Network Design
Spaces](https://arxiv.org/abs/2003.13678)
The main idea is to start with a high dimensional search space and
iteratively reduce the search space by empirically apply constrains
based on the best performing models sampled by the current search
space.
The resulting models are light, accurate, and faster than
EfficientNets (up to 5x times!)
For example, to go from $AnyNet_A$ to $AnyNet_B$ they fixed the
bottleneck ratio $b_i$ for all stage $i$. The following table shows
all the restrictions applied from one search space to the next one.
The paper is really well written and very interesting, I highly
recommended read it.
``` python
ResNet.regnetx_002()
ResNet.regnetx_004()
ResNet.regnetx_006()
ResNet.regnetx_008()
ResNet.regnetx_016()
ResNet.regnetx_040()
ResNet.regnetx_064()
ResNet.regnetx_080()
ResNet.regnetx_120()
ResNet.regnetx_160()
ResNet.regnetx_320()
# Y variants (with SE)
ResNet.regnety_002()
# ...
ResNet.regnetx_320()
You can easily customize your model
```
Examples:
``` python
# change activation
RegNet.regnetx_004(activation = nn.SELU)
# change number of classes (default is 1000 )
RegNet.regnetx_004(n_classes=100)
# pass a different block
RegNet.regnetx_004(block=RegNetYBotteneckBlock)
# change the steam
model = RegNet.regnetx_004(stem=ResNetStemC)
change shortcut
model = RegNet.regnetx_004(block=partial(RegNetYBotteneckBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = RegNet.regnetx_004()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 32, 112, 112]), torch.Size([1, 32, 56, 56]), torch.Size([1, 64, 28, 28]), torch.Size([1, 160, 14, 14])]
```
|
|
glasses/regnetx_032 | 2021-04-21T19:07:57.000Z | [
"pytorch",
"arxiv:2003.13678",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 7 | transformers | # regnetx_032
Implementation of RegNet proposed in [Designing Network Design
Spaces](https://arxiv.org/abs/2003.13678)
The main idea is to start with a high dimensional search space and
iteratively reduce the search space by empirically apply constrains
based on the best performing models sampled by the current search
space.
The resulting models are light, accurate, and faster than
EfficientNets (up to 5x times!)
For example, to go from $AnyNet_A$ to $AnyNet_B$ they fixed the
bottleneck ratio $b_i$ for all stage $i$. The following table shows
all the restrictions applied from one search space to the next one.
The paper is really well written and very interesting, I highly
recommended read it.
``` python
ResNet.regnetx_002()
ResNet.regnetx_004()
ResNet.regnetx_006()
ResNet.regnetx_008()
ResNet.regnetx_016()
ResNet.regnetx_040()
ResNet.regnetx_064()
ResNet.regnetx_080()
ResNet.regnetx_120()
ResNet.regnetx_160()
ResNet.regnetx_320()
# Y variants (with SE)
ResNet.regnety_002()
# ...
ResNet.regnetx_320()
You can easily customize your model
```
Examples:
``` python
# change activation
RegNet.regnetx_004(activation = nn.SELU)
# change number of classes (default is 1000 )
RegNet.regnetx_004(n_classes=100)
# pass a different block
RegNet.regnetx_004(block=RegNetYBotteneckBlock)
# change the steam
model = RegNet.regnetx_004(stem=ResNetStemC)
change shortcut
model = RegNet.regnetx_004(block=partial(RegNetYBotteneckBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = RegNet.regnetx_004()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 32, 112, 112]), torch.Size([1, 32, 56, 56]), torch.Size([1, 64, 28, 28]), torch.Size([1, 160, 14, 14])]
```
|
|
glasses/regnetx_040 | 2021-04-24T10:54:15.000Z | [
"pytorch",
"transformers"
]
| [
".gitattributes",
"config.json",
"pytorch_model.bin"
]
| glasses | 7 | transformers | ||
glasses/regnetx_064 | 2021-04-25T07:57:31.000Z | [
"pytorch",
"transformers"
]
| [
".gitattributes",
"config.json",
"pytorch_model.bin"
]
| glasses | 8 | transformers | ||
glasses/regnety_002 | 2021-04-21T19:08:02.000Z | [
"pytorch",
"arxiv:2003.13678",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 11 | transformers | # regnety_002
Implementation of RegNet proposed in [Designing Network Design
Spaces](https://arxiv.org/abs/2003.13678)
The main idea is to start with a high dimensional search space and
iteratively reduce the search space by empirically apply constrains
based on the best performing models sampled by the current search
space.
The resulting models are light, accurate, and faster than
EfficientNets (up to 5x times!)
For example, to go from $AnyNet_A$ to $AnyNet_B$ they fixed the
bottleneck ratio $b_i$ for all stage $i$. The following table shows
all the restrictions applied from one search space to the next one.
The paper is really well written and very interesting, I highly
recommended read it.
``` python
ResNet.regnetx_002()
ResNet.regnetx_004()
ResNet.regnetx_006()
ResNet.regnetx_008()
ResNet.regnetx_016()
ResNet.regnetx_040()
ResNet.regnetx_064()
ResNet.regnetx_080()
ResNet.regnetx_120()
ResNet.regnetx_160()
ResNet.regnetx_320()
# Y variants (with SE)
ResNet.regnety_002()
# ...
ResNet.regnetx_320()
You can easily customize your model
```
Examples:
``` python
# change activation
RegNet.regnetx_004(activation = nn.SELU)
# change number of classes (default is 1000 )
RegNet.regnetx_004(n_classes=100)
# pass a different block
RegNet.regnetx_004(block=RegNetYBotteneckBlock)
# change the steam
model = RegNet.regnetx_004(stem=ResNetStemC)
change shortcut
model = RegNet.regnetx_004(block=partial(RegNetYBotteneckBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = RegNet.regnetx_004()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 32, 112, 112]), torch.Size([1, 32, 56, 56]), torch.Size([1, 64, 28, 28]), torch.Size([1, 160, 14, 14])]
```
|
|
glasses/regnety_004 | 2021-04-21T19:08:07.000Z | [
"pytorch",
"arxiv:2003.13678",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 6 | transformers | # regnety_004
Implementation of RegNet proposed in [Designing Network Design
Spaces](https://arxiv.org/abs/2003.13678)
The main idea is to start with a high dimensional search space and
iteratively reduce the search space by empirically apply constrains
based on the best performing models sampled by the current search
space.
The resulting models are light, accurate, and faster than
EfficientNets (up to 5x times!)
For example, to go from $AnyNet_A$ to $AnyNet_B$ they fixed the
bottleneck ratio $b_i$ for all stage $i$. The following table shows
all the restrictions applied from one search space to the next one.
The paper is really well written and very interesting, I highly
recommended read it.
``` python
ResNet.regnetx_002()
ResNet.regnetx_004()
ResNet.regnetx_006()
ResNet.regnetx_008()
ResNet.regnetx_016()
ResNet.regnetx_040()
ResNet.regnetx_064()
ResNet.regnetx_080()
ResNet.regnetx_120()
ResNet.regnetx_160()
ResNet.regnetx_320()
# Y variants (with SE)
ResNet.regnety_002()
# ...
ResNet.regnetx_320()
You can easily customize your model
```
Examples:
``` python
# change activation
RegNet.regnetx_004(activation = nn.SELU)
# change number of classes (default is 1000 )
RegNet.regnetx_004(n_classes=100)
# pass a different block
RegNet.regnetx_004(block=RegNetYBotteneckBlock)
# change the steam
model = RegNet.regnetx_004(stem=ResNetStemC)
change shortcut
model = RegNet.regnetx_004(block=partial(RegNetYBotteneckBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = RegNet.regnetx_004()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 32, 112, 112]), torch.Size([1, 32, 56, 56]), torch.Size([1, 64, 28, 28]), torch.Size([1, 160, 14, 14])]
```
|
|
glasses/regnety_006 | 2021-04-21T19:08:12.000Z | [
"pytorch",
"arxiv:2003.13678",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 7 | transformers | # regnety_006
Implementation of RegNet proposed in [Designing Network Design
Spaces](https://arxiv.org/abs/2003.13678)
The main idea is to start with a high dimensional search space and
iteratively reduce the search space by empirically apply constrains
based on the best performing models sampled by the current search
space.
The resulting models are light, accurate, and faster than
EfficientNets (up to 5x times!)
For example, to go from $AnyNet_A$ to $AnyNet_B$ they fixed the
bottleneck ratio $b_i$ for all stage $i$. The following table shows
all the restrictions applied from one search space to the next one.
The paper is really well written and very interesting, I highly
recommended read it.
``` python
ResNet.regnetx_002()
ResNet.regnetx_004()
ResNet.regnetx_006()
ResNet.regnetx_008()
ResNet.regnetx_016()
ResNet.regnetx_040()
ResNet.regnetx_064()
ResNet.regnetx_080()
ResNet.regnetx_120()
ResNet.regnetx_160()
ResNet.regnetx_320()
# Y variants (with SE)
ResNet.regnety_002()
# ...
ResNet.regnetx_320()
You can easily customize your model
```
Examples:
``` python
# change activation
RegNet.regnetx_004(activation = nn.SELU)
# change number of classes (default is 1000 )
RegNet.regnetx_004(n_classes=100)
# pass a different block
RegNet.regnetx_004(block=RegNetYBotteneckBlock)
# change the steam
model = RegNet.regnetx_004(stem=ResNetStemC)
change shortcut
model = RegNet.regnetx_004(block=partial(RegNetYBotteneckBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = RegNet.regnetx_004()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 32, 112, 112]), torch.Size([1, 32, 56, 56]), torch.Size([1, 64, 28, 28]), torch.Size([1, 160, 14, 14])]
```
|
|
glasses/regnety_008 | 2021-04-21T19:08:18.000Z | [
"pytorch",
"arxiv:2003.13678",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 7 | transformers | # regnety_008
Implementation of RegNet proposed in [Designing Network Design
Spaces](https://arxiv.org/abs/2003.13678)
The main idea is to start with a high dimensional search space and
iteratively reduce the search space by empirically apply constrains
based on the best performing models sampled by the current search
space.
The resulting models are light, accurate, and faster than
EfficientNets (up to 5x times!)
For example, to go from $AnyNet_A$ to $AnyNet_B$ they fixed the
bottleneck ratio $b_i$ for all stage $i$. The following table shows
all the restrictions applied from one search space to the next one.
The paper is really well written and very interesting, I highly
recommended read it.
``` python
ResNet.regnetx_002()
ResNet.regnetx_004()
ResNet.regnetx_006()
ResNet.regnetx_008()
ResNet.regnetx_016()
ResNet.regnetx_040()
ResNet.regnetx_064()
ResNet.regnetx_080()
ResNet.regnetx_120()
ResNet.regnetx_160()
ResNet.regnetx_320()
# Y variants (with SE)
ResNet.regnety_002()
# ...
ResNet.regnetx_320()
You can easily customize your model
```
Examples:
``` python
# change activation
RegNet.regnetx_004(activation = nn.SELU)
# change number of classes (default is 1000 )
RegNet.regnetx_004(n_classes=100)
# pass a different block
RegNet.regnetx_004(block=RegNetYBotteneckBlock)
# change the steam
model = RegNet.regnetx_004(stem=ResNetStemC)
change shortcut
model = RegNet.regnetx_004(block=partial(RegNetYBotteneckBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = RegNet.regnetx_004()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 32, 112, 112]), torch.Size([1, 32, 56, 56]), torch.Size([1, 64, 28, 28]), torch.Size([1, 160, 14, 14])]
```
|
|
glasses/regnety_016 | 2021-04-21T19:08:24.000Z | [
"pytorch",
"arxiv:2003.13678",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 9 | transformers | # regnety_016
Implementation of RegNet proposed in [Designing Network Design
Spaces](https://arxiv.org/abs/2003.13678)
The main idea is to start with a high dimensional search space and
iteratively reduce the search space by empirically apply constrains
based on the best performing models sampled by the current search
space.
The resulting models are light, accurate, and faster than
EfficientNets (up to 5x times!)
For example, to go from $AnyNet_A$ to $AnyNet_B$ they fixed the
bottleneck ratio $b_i$ for all stage $i$. The following table shows
all the restrictions applied from one search space to the next one.
The paper is really well written and very interesting, I highly
recommended read it.
``` python
ResNet.regnetx_002()
ResNet.regnetx_004()
ResNet.regnetx_006()
ResNet.regnetx_008()
ResNet.regnetx_016()
ResNet.regnetx_040()
ResNet.regnetx_064()
ResNet.regnetx_080()
ResNet.regnetx_120()
ResNet.regnetx_160()
ResNet.regnetx_320()
# Y variants (with SE)
ResNet.regnety_002()
# ...
ResNet.regnetx_320()
You can easily customize your model
```
Examples:
``` python
# change activation
RegNet.regnetx_004(activation = nn.SELU)
# change number of classes (default is 1000 )
RegNet.regnetx_004(n_classes=100)
# pass a different block
RegNet.regnetx_004(block=RegNetYBotteneckBlock)
# change the steam
model = RegNet.regnetx_004(stem=ResNetStemC)
change shortcut
model = RegNet.regnetx_004(block=partial(RegNetYBotteneckBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = RegNet.regnetx_004()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 32, 112, 112]), torch.Size([1, 32, 56, 56]), torch.Size([1, 64, 28, 28]), torch.Size([1, 160, 14, 14])]
```
|
|
glasses/regnety_032 | 2021-04-21T19:08:31.000Z | [
"pytorch",
"arxiv:2003.13678",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 7 | transformers | # regnety_032
Implementation of RegNet proposed in [Designing Network Design
Spaces](https://arxiv.org/abs/2003.13678)
The main idea is to start with a high dimensional search space and
iteratively reduce the search space by empirically apply constrains
based on the best performing models sampled by the current search
space.
The resulting models are light, accurate, and faster than
EfficientNets (up to 5x times!)
For example, to go from $AnyNet_A$ to $AnyNet_B$ they fixed the
bottleneck ratio $b_i$ for all stage $i$. The following table shows
all the restrictions applied from one search space to the next one.
The paper is really well written and very interesting, I highly
recommended read it.
``` python
ResNet.regnetx_002()
ResNet.regnetx_004()
ResNet.regnetx_006()
ResNet.regnetx_008()
ResNet.regnetx_016()
ResNet.regnetx_040()
ResNet.regnetx_064()
ResNet.regnetx_080()
ResNet.regnetx_120()
ResNet.regnetx_160()
ResNet.regnetx_320()
# Y variants (with SE)
ResNet.regnety_002()
# ...
ResNet.regnetx_320()
You can easily customize your model
```
Examples:
``` python
# change activation
RegNet.regnetx_004(activation = nn.SELU)
# change number of classes (default is 1000 )
RegNet.regnetx_004(n_classes=100)
# pass a different block
RegNet.regnetx_004(block=RegNetYBotteneckBlock)
# change the steam
model = RegNet.regnetx_004(stem=ResNetStemC)
change shortcut
model = RegNet.regnetx_004(block=partial(RegNetYBotteneckBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = RegNet.regnetx_004()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 32, 112, 112]), torch.Size([1, 32, 56, 56]), torch.Size([1, 64, 28, 28]), torch.Size([1, 160, 14, 14])]
```
|
|
glasses/regnety_040 | 2021-04-25T07:58:31.000Z | [
"pytorch",
"transformers"
]
| [
".gitattributes",
"config.json",
"pytorch_model.bin"
]
| glasses | 10 | transformers | ||
glasses/regnety_064 | 2021-04-25T07:58:47.000Z | [
"pytorch",
"transformers"
]
| [
".gitattributes",
"config.json",
"pytorch_model.bin"
]
| glasses | 10 | transformers | ||
glasses/resnet101 | 2021-04-24T10:50:15.000Z | [
"pytorch",
"arxiv:1512.03385",
"arxiv:1812.01187",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 12 | transformers | # resnet101
Implementation of ResNet proposed in [Deep Residual Learning for Image
Recognition](https://arxiv.org/abs/1512.03385)
``` python
ResNet.resnet18()
ResNet.resnet26()
ResNet.resnet34()
ResNet.resnet50()
ResNet.resnet101()
ResNet.resnet152()
ResNet.resnet200()
Variants (d) proposed in `Bag of Tricks for Image Classification with Convolutional Neural Networks <https://arxiv.org/pdf/1812.01187.pdf`_
ResNet.resnet26d()
ResNet.resnet34d()
ResNet.resnet50d()
# You can construct your own one by chaning `stem` and `block`
resnet101d = ResNet.resnet101(stem=ResNetStemC, block=partial(ResNetBottleneckBlock, shortcut=ResNetShorcutD))
```
Examples:
``` python
# change activation
ResNet.resnet18(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNet.resnet18(n_classes=100)
# pass a different block
ResNet.resnet18(block=SENetBasicBlock)
# change the steam
model = ResNet.resnet18(stem=ResNetStemC)
change shortcut
model = ResNet.resnet18(block=partial(ResNetBasicBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = ResNet.resnet18()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
|
glasses/resnet152 | 2021-04-24T10:50:39.000Z | [
"pytorch",
"arxiv:1512.03385",
"arxiv:1812.01187",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 10 | transformers | # resnet152
Implementation of ResNet proposed in [Deep Residual Learning for Image
Recognition](https://arxiv.org/abs/1512.03385)
``` python
ResNet.resnet18()
ResNet.resnet26()
ResNet.resnet34()
ResNet.resnet50()
ResNet.resnet101()
ResNet.resnet152()
ResNet.resnet200()
Variants (d) proposed in `Bag of Tricks for Image Classification with Convolutional Neural Networks <https://arxiv.org/pdf/1812.01187.pdf`_
ResNet.resnet26d()
ResNet.resnet34d()
ResNet.resnet50d()
# You can construct your own one by chaning `stem` and `block`
resnet101d = ResNet.resnet101(stem=ResNetStemC, block=partial(ResNetBottleneckBlock, shortcut=ResNetShorcutD))
```
Examples:
``` python
# change activation
ResNet.resnet18(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNet.resnet18(n_classes=100)
# pass a different block
ResNet.resnet18(block=SENetBasicBlock)
# change the steam
model = ResNet.resnet18(stem=ResNetStemC)
change shortcut
model = ResNet.resnet18(block=partial(ResNetBasicBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = ResNet.resnet18()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
|
glasses/resnet18 | 2021-04-24T10:48:45.000Z | [
"pytorch",
"arxiv:1512.03385",
"arxiv:1812.01187",
"transformers"
]
| [
".gitattributes",
".org.chromium.Chromium.DyPFg4",
"README.md",
"config-err-Edzi90",
"config.json",
"pytorch_model.bin",
"vscode-ipc",
"VSFeedbackVSRTCLogs/20210409_083938_16179575780260_VSCode.log",
"VSFeedbackVSRTCLogs/20210409_083938_Agent.log",
"VSFeedbackVSRTCLogs/20210409_103612_16179645728160_VSCode.log",
"VSFeedbackVSRTCLogs/20210409_103613_Agent.log",
"VSFeedbackVSRTCLogs/20210409_110958_16179665980940_VSCode.log",
"remote-file-71752ccfe2464e99/1d68001ef112fa197e0de1f2cbc5ab57ebf464a5ae676667c4e51b82762627a7/0178b70fdabe-0x8D8F230C59F9D48",
"remote-file-71752ccfe2464e99/1d68001ef112fa197e0de1f2cbc5ab57ebf464a5ae676667c4e51b82762627a7/0178b77db7fd-0x8D8F230C59F9D48",
"resnet18/config.json",
"resnet18/pytorch_model.bin"
]
| glasses | 25 | transformers | # resnet18
Implementation of ResNet proposed in [Deep Residual Learning for Image
Recognition](https://arxiv.org/abs/1512.03385)
``` python
ResNet.resnet18()
ResNet.resnet26()
ResNet.resnet34()
ResNet.resnet50()
ResNet.resnet101()
ResNet.resnet152()
ResNet.resnet200()
Variants (d) proposed in `Bag of Tricks for Image Classification with Convolutional Neural Networks <https://arxiv.org/pdf/1812.01187.pdf`_
ResNet.resnet26d()
ResNet.resnet34d()
ResNet.resnet50d()
# You can construct your own one by chaning `stem` and `block`
resnet101d = ResNet.resnet101(stem=ResNetStemC, block=partial(ResNetBottleneckBlock, shortcut=ResNetShorcutD))
```
Examples:
``` python
# change activation
ResNet.resnet18(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNet.resnet18(n_classes=100)
# pass a different block
ResNet.resnet18(block=SENetBasicBlock)
# change the steam
model = ResNet.resnet18(stem=ResNetStemC)
change shortcut
model = ResNet.resnet18(block=partial(ResNetBasicBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = ResNet.resnet18()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
|
glasses/resnet26 | 2021-04-24T10:48:55.000Z | [
"pytorch",
"arxiv:1512.03385",
"arxiv:1812.01187",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 8 | transformers | # resnet26
Implementation of ResNet proposed in [Deep Residual Learning for Image
Recognition](https://arxiv.org/abs/1512.03385)
``` python
ResNet.resnet18()
ResNet.resnet26()
ResNet.resnet34()
ResNet.resnet50()
ResNet.resnet101()
ResNet.resnet152()
ResNet.resnet200()
Variants (d) proposed in `Bag of Tricks for Image Classification with Convolutional Neural Networks <https://arxiv.org/pdf/1812.01187.pdf`_
ResNet.resnet26d()
ResNet.resnet34d()
ResNet.resnet50d()
# You can construct your own one by chaning `stem` and `block`
resnet101d = ResNet.resnet101(stem=ResNetStemC, block=partial(ResNetBottleneckBlock, shortcut=ResNetShorcutD))
```
Examples:
``` python
# change activation
ResNet.resnet18(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNet.resnet18(n_classes=100)
# pass a different block
ResNet.resnet18(block=SENetBasicBlock)
# change the steam
model = ResNet.resnet18(stem=ResNetStemC)
change shortcut
model = ResNet.resnet18(block=partial(ResNetBasicBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = ResNet.resnet18()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
|
glasses/resnet26d | 2021-04-24T10:49:05.000Z | [
"pytorch",
"arxiv:1512.03385",
"arxiv:1812.01187",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 9 | transformers | # resnet26d
Implementation of ResNet proposed in [Deep Residual Learning for Image
Recognition](https://arxiv.org/abs/1512.03385)
``` python
ResNet.resnet18()
ResNet.resnet26()
ResNet.resnet34()
ResNet.resnet50()
ResNet.resnet101()
ResNet.resnet152()
ResNet.resnet200()
Variants (d) proposed in `Bag of Tricks for Image Classification with Convolutional Neural Networks <https://arxiv.org/pdf/1812.01187.pdf`_
ResNet.resnet26d()
ResNet.resnet34d()
ResNet.resnet50d()
# You can construct your own one by chaning `stem` and `block`
resnet101d = ResNet.resnet101(stem=ResNetStemC, block=partial(ResNetBottleneckBlock, shortcut=ResNetShorcutD))
```
Examples:
``` python
# change activation
ResNet.resnet18(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNet.resnet18(n_classes=100)
# pass a different block
ResNet.resnet18(block=SENetBasicBlock)
# change the steam
model = ResNet.resnet18(stem=ResNetStemC)
change shortcut
model = ResNet.resnet18(block=partial(ResNetBasicBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = ResNet.resnet18()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
|
glasses/resnet34 | 2021-04-24T10:49:20.000Z | [
"pytorch",
"arxiv:1512.03385",
"arxiv:1812.01187",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 10 | transformers | # resnet34
Implementation of ResNet proposed in [Deep Residual Learning for Image
Recognition](https://arxiv.org/abs/1512.03385)
``` python
ResNet.resnet18()
ResNet.resnet26()
ResNet.resnet34()
ResNet.resnet50()
ResNet.resnet101()
ResNet.resnet152()
ResNet.resnet200()
Variants (d) proposed in `Bag of Tricks for Image Classification with Convolutional Neural Networks <https://arxiv.org/pdf/1812.01187.pdf`_
ResNet.resnet26d()
ResNet.resnet34d()
ResNet.resnet50d()
# You can construct your own one by chaning `stem` and `block`
resnet101d = ResNet.resnet101(stem=ResNetStemC, block=partial(ResNetBottleneckBlock, shortcut=ResNetShorcutD))
```
Examples:
``` python
# change activation
ResNet.resnet18(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNet.resnet18(n_classes=100)
# pass a different block
ResNet.resnet18(block=SENetBasicBlock)
# change the steam
model = ResNet.resnet18(stem=ResNetStemC)
change shortcut
model = ResNet.resnet18(block=partial(ResNetBasicBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = ResNet.resnet18()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
|
glasses/resnet34d | 2021-04-24T10:49:30.000Z | [
"pytorch",
"arxiv:1512.03385",
"arxiv:1812.01187",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 7 | transformers | # resnet34d
Implementation of ResNet proposed in [Deep Residual Learning for Image
Recognition](https://arxiv.org/abs/1512.03385)
``` python
ResNet.resnet18()
ResNet.resnet26()
ResNet.resnet34()
ResNet.resnet50()
ResNet.resnet101()
ResNet.resnet152()
ResNet.resnet200()
Variants (d) proposed in `Bag of Tricks for Image Classification with Convolutional Neural Networks <https://arxiv.org/pdf/1812.01187.pdf`_
ResNet.resnet26d()
ResNet.resnet34d()
ResNet.resnet50d()
# You can construct your own one by chaning `stem` and `block`
resnet101d = ResNet.resnet101(stem=ResNetStemC, block=partial(ResNetBottleneckBlock, shortcut=ResNetShorcutD))
```
Examples:
``` python
# change activation
ResNet.resnet18(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNet.resnet18(n_classes=100)
# pass a different block
ResNet.resnet18(block=SENetBasicBlock)
# change the steam
model = ResNet.resnet18(stem=ResNetStemC)
change shortcut
model = ResNet.resnet18(block=partial(ResNetBasicBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = ResNet.resnet18()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
|
glasses/resnet50 | 2021-04-24T10:49:43.000Z | [
"pytorch",
"arxiv:1512.03385",
"arxiv:1812.01187",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 12 | transformers | # resnet50
Implementation of ResNet proposed in [Deep Residual Learning for Image
Recognition](https://arxiv.org/abs/1512.03385)
``` python
ResNet.resnet18()
ResNet.resnet26()
ResNet.resnet34()
ResNet.resnet50()
ResNet.resnet101()
ResNet.resnet152()
ResNet.resnet200()
Variants (d) proposed in `Bag of Tricks for Image Classification with Convolutional Neural Networks <https://arxiv.org/pdf/1812.01187.pdf`_
ResNet.resnet26d()
ResNet.resnet34d()
ResNet.resnet50d()
# You can construct your own one by chaning `stem` and `block`
resnet101d = ResNet.resnet101(stem=ResNetStemC, block=partial(ResNetBottleneckBlock, shortcut=ResNetShorcutD))
```
Examples:
``` python
# change activation
ResNet.resnet18(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNet.resnet18(n_classes=100)
# pass a different block
ResNet.resnet18(block=SENetBasicBlock)
# change the steam
model = ResNet.resnet18(stem=ResNetStemC)
change shortcut
model = ResNet.resnet18(block=partial(ResNetBasicBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = ResNet.resnet18()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
|
glasses/resnet50d | 2021-04-24T10:49:57.000Z | [
"pytorch",
"arxiv:1512.03385",
"arxiv:1812.01187",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 10 | transformers | # resnet50d
Implementation of ResNet proposed in [Deep Residual Learning for Image
Recognition](https://arxiv.org/abs/1512.03385)
``` python
ResNet.resnet18()
ResNet.resnet26()
ResNet.resnet34()
ResNet.resnet50()
ResNet.resnet101()
ResNet.resnet152()
ResNet.resnet200()
Variants (d) proposed in `Bag of Tricks for Image Classification with Convolutional Neural Networks <https://arxiv.org/pdf/1812.01187.pdf`_
ResNet.resnet26d()
ResNet.resnet34d()
ResNet.resnet50d()
# You can construct your own one by chaning `stem` and `block`
resnet101d = ResNet.resnet101(stem=ResNetStemC, block=partial(ResNetBottleneckBlock, shortcut=ResNetShorcutD))
```
Examples:
``` python
# change activation
ResNet.resnet18(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNet.resnet18(n_classes=100)
# pass a different block
ResNet.resnet18(block=SENetBasicBlock)
# change the steam
model = ResNet.resnet18(stem=ResNetStemC)
change shortcut
model = ResNet.resnet18(block=partial(ResNetBasicBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = ResNet.resnet18()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
|
glasses/resnext101_32x8d | 2021-04-24T10:51:37.000Z | [
"pytorch",
"arxiv:1611.05431",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 7 | transformers | # resnext101_32x8d
Implementation of ResNetXt proposed in [\"Aggregated Residual
Transformation for Deep Neural
Networks\"](https://arxiv.org/pdf/1611.05431.pdf)
Create a default model
``` python
ResNetXt.resnext50_32x4d()
ResNetXt.resnext101_32x8d()
# create a resnetxt18_32x4d
ResNetXt.resnet18(block=ResNetXtBottleNeckBlock, groups=32, base_width=4)
```
Examples:
: ``` python
# change activation
ResNetXt.resnext50_32x4d(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNetXt.resnext50_32x4d(n_classes=100)
# pass a different block
ResNetXt.resnext50_32x4d(block=SENetBasicBlock)
# change the initial convolution
model = ResNetXt.resnext50_32x4d
model.encoder.gate.conv1 = nn.Conv2d(3, 64, kernel_size=3)
# store each feature
x = torch.rand((1, 3, 224, 224))
model = ResNetXt.resnext50_32x4d()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
|
glasses/resnext50_32x4d | 2021-04-24T10:51:12.000Z | [
"pytorch",
"arxiv:1611.05431",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 13 | transformers | # resnext50_32x4d
Implementation of ResNetXt proposed in [\"Aggregated Residual
Transformation for Deep Neural
Networks\"](https://arxiv.org/pdf/1611.05431.pdf)
Create a default model
``` python
ResNetXt.resnext50_32x4d()
ResNetXt.resnext101_32x8d()
# create a resnetxt18_32x4d
ResNetXt.resnet18(block=ResNetXtBottleNeckBlock, groups=32, base_width=4)
```
Examples:
: ``` python
# change activation
ResNetXt.resnext50_32x4d(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNetXt.resnext50_32x4d(n_classes=100)
# pass a different block
ResNetXt.resnext50_32x4d(block=SENetBasicBlock)
# change the initial convolution
model = ResNetXt.resnext50_32x4d
model.encoder.gate.conv1 = nn.Conv2d(3, 64, kernel_size=3)
# store each feature
x = torch.rand((1, 3, 224, 224))
model = ResNetXt.resnext50_32x4d()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
|
glasses/vgg11 | 2021-04-21T19:10:34.000Z | [
"pytorch",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 8 | transformers | # vgg11
Implementation of VGG proposed in [Very Deep Convolutional Networks For
Large-Scale Image Recognition](https://arxiv.org/pdf/1409.1556.pdf)
``` python
VGG.vgg11()
VGG.vgg13()
VGG.vgg16()
VGG.vgg19()
VGG.vgg11_bn()
VGG.vgg13_bn()
VGG.vgg16_bn()
VGG.vgg19_bn()
```
Please be aware that the [bn]{.title-ref} models uses BatchNorm but
they are very old and people back then don\'t know the bias is
superfluous in a conv followed by a batchnorm.
Examples:
``` python
# change activation
VGG.vgg11(activation = nn.SELU)
# change number of classes (default is 1000 )
VGG.vgg11(n_classes=100)
# pass a different block
from nn.models.classification.senet import SENetBasicBlock
VGG.vgg11(block=SENetBasicBlock)
# store the features tensor after every block
```
|
|
glasses/vgg11_bn | 2021-04-22T17:58:21.000Z | [
"pytorch",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 9 | transformers | # vgg11_bn
Implementation of VGG proposed in [Very Deep Convolutional Networks For
Large-Scale Image Recognition](https://arxiv.org/pdf/1409.1556.pdf)
``` python
VGG.vgg11()
VGG.vgg13()
VGG.vgg16()
VGG.vgg19()
VGG.vgg11_bn()
VGG.vgg13_bn()
VGG.vgg16_bn()
VGG.vgg19_bn()
```
Please be aware that the [bn]{.title-ref} models uses BatchNorm but
they are very old and people back then don\'t know the bias is
superfluous in a conv followed by a batchnorm.
Examples:
``` python
# change activation
VGG.vgg11(activation = nn.SELU)
# change number of classes (default is 1000 )
VGG.vgg11(n_classes=100)
# pass a different block
from nn.models.classification.senet import SENetBasicBlock
VGG.vgg11(block=SENetBasicBlock)
# store the features tensor after every block
```
|
|
glasses/vgg13 | 2021-04-21T19:10:58.000Z | [
"pytorch",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 14 | transformers | # vgg13
Implementation of VGG proposed in [Very Deep Convolutional Networks For
Large-Scale Image Recognition](https://arxiv.org/pdf/1409.1556.pdf)
``` python
VGG.vgg11()
VGG.vgg13()
VGG.vgg16()
VGG.vgg19()
VGG.vgg11_bn()
VGG.vgg13_bn()
VGG.vgg16_bn()
VGG.vgg19_bn()
```
Please be aware that the [bn]{.title-ref} models uses BatchNorm but
they are very old and people back then don\'t know the bias is
superfluous in a conv followed by a batchnorm.
Examples:
``` python
# change activation
VGG.vgg11(activation = nn.SELU)
# change number of classes (default is 1000 )
VGG.vgg11(n_classes=100)
# pass a different block
from nn.models.classification.senet import SENetBasicBlock
VGG.vgg11(block=SENetBasicBlock)
# store the features tensor after every block
```
|
|
glasses/vgg13_bn | 2021-04-22T17:58:45.000Z | [
"pytorch",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 8 | transformers | # vgg13_bn
Implementation of VGG proposed in [Very Deep Convolutional Networks For
Large-Scale Image Recognition](https://arxiv.org/pdf/1409.1556.pdf)
``` python
VGG.vgg11()
VGG.vgg13()
VGG.vgg16()
VGG.vgg19()
VGG.vgg11_bn()
VGG.vgg13_bn()
VGG.vgg16_bn()
VGG.vgg19_bn()
```
Please be aware that the [bn]{.title-ref} models uses BatchNorm but
they are very old and people back then don\'t know the bias is
superfluous in a conv followed by a batchnorm.
Examples:
``` python
# change activation
VGG.vgg11(activation = nn.SELU)
# change number of classes (default is 1000 )
VGG.vgg11(n_classes=100)
# pass a different block
from nn.models.classification.senet import SENetBasicBlock
VGG.vgg11(block=SENetBasicBlock)
# store the features tensor after every block
```
|
|
glasses/vgg16 | 2021-04-21T19:11:22.000Z | [
"pytorch",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 7 | transformers | # vgg16
Implementation of VGG proposed in [Very Deep Convolutional Networks For
Large-Scale Image Recognition](https://arxiv.org/pdf/1409.1556.pdf)
``` python
VGG.vgg11()
VGG.vgg13()
VGG.vgg16()
VGG.vgg19()
VGG.vgg11_bn()
VGG.vgg13_bn()
VGG.vgg16_bn()
VGG.vgg19_bn()
```
Please be aware that the [bn]{.title-ref} models uses BatchNorm but
they are very old and people back then don\'t know the bias is
superfluous in a conv followed by a batchnorm.
Examples:
``` python
# change activation
VGG.vgg11(activation = nn.SELU)
# change number of classes (default is 1000 )
VGG.vgg11(n_classes=100)
# pass a different block
from nn.models.classification.senet import SENetBasicBlock
VGG.vgg11(block=SENetBasicBlock)
# store the features tensor after every block
```
|
|
glasses/vgg16_bn | 2021-04-22T17:59:10.000Z | [
"pytorch",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 8 | transformers | # vgg16_bn
Implementation of VGG proposed in [Very Deep Convolutional Networks For
Large-Scale Image Recognition](https://arxiv.org/pdf/1409.1556.pdf)
``` python
VGG.vgg11()
VGG.vgg13()
VGG.vgg16()
VGG.vgg19()
VGG.vgg11_bn()
VGG.vgg13_bn()
VGG.vgg16_bn()
VGG.vgg19_bn()
```
Please be aware that the [bn]{.title-ref} models uses BatchNorm but
they are very old and people back then don\'t know the bias is
superfluous in a conv followed by a batchnorm.
Examples:
``` python
# change activation
VGG.vgg11(activation = nn.SELU)
# change number of classes (default is 1000 )
VGG.vgg11(n_classes=100)
# pass a different block
from nn.models.classification.senet import SENetBasicBlock
VGG.vgg11(block=SENetBasicBlock)
# store the features tensor after every block
```
|
|
glasses/vgg19 | 2021-04-22T17:58:01.000Z | [
"pytorch",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 12 | transformers | # vgg19
Implementation of VGG proposed in [Very Deep Convolutional Networks For
Large-Scale Image Recognition](https://arxiv.org/pdf/1409.1556.pdf)
``` python
VGG.vgg11()
VGG.vgg13()
VGG.vgg16()
VGG.vgg19()
VGG.vgg11_bn()
VGG.vgg13_bn()
VGG.vgg16_bn()
VGG.vgg19_bn()
```
Please be aware that the [bn]{.title-ref} models uses BatchNorm but
they are very old and people back then don\'t know the bias is
superfluous in a conv followed by a batchnorm.
Examples:
``` python
# change activation
VGG.vgg11(activation = nn.SELU)
# change number of classes (default is 1000 )
VGG.vgg11(n_classes=100)
# pass a different block
from nn.models.classification.senet import SENetBasicBlock
VGG.vgg11(block=SENetBasicBlock)
# store the features tensor after every block
```
|
|
glasses/vgg19_bn | 2021-04-22T17:59:37.000Z | [
"pytorch",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 7 | transformers | # vgg19_bn
Implementation of VGG proposed in [Very Deep Convolutional Networks For
Large-Scale Image Recognition](https://arxiv.org/pdf/1409.1556.pdf)
``` python
VGG.vgg11()
VGG.vgg13()
VGG.vgg16()
VGG.vgg19()
VGG.vgg11_bn()
VGG.vgg13_bn()
VGG.vgg16_bn()
VGG.vgg19_bn()
```
Please be aware that the [bn]{.title-ref} models uses BatchNorm but
they are very old and people back then don\'t know the bias is
superfluous in a conv followed by a batchnorm.
Examples:
``` python
# change activation
VGG.vgg11(activation = nn.SELU)
# change number of classes (default is 1000 )
VGG.vgg11(n_classes=100)
# pass a different block
from nn.models.classification.senet import SENetBasicBlock
VGG.vgg11(block=SENetBasicBlock)
# store the features tensor after every block
```
|
|
glasses/vit_base_patch16_224 | 2021-04-22T18:37:04.000Z | [
"pytorch",
"arxiv:2010.11929",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 28 | transformers | # vit_base_patch16_224
Implementation of Vision Transformer (ViT) proposed in [An Image Is
Worth 16x16 Words: Transformers For Image Recognition At
Scale](https://arxiv.org/pdf/2010.11929.pdf)
The following image from the authors shows the architecture.
``` python
ViT.vit_small_patch16_224()
ViT.vit_base_patch16_224()
ViT.vit_base_patch16_384()
ViT.vit_base_patch32_384()
ViT.vit_huge_patch16_224()
ViT.vit_huge_patch32_384()
ViT.vit_large_patch16_224()
ViT.vit_large_patch16_384()
ViT.vit_large_patch32_384()
```
Examples:
``` python
# change activation
ViT.vit_base_patch16_224(activation = nn.SELU)
# change number of classes (default is 1000 )
ViT.vit_base_patch16_224(n_classes=100)
# pass a different block, default is TransformerEncoderBlock
ViT.vit_base_patch16_224(block=MyCoolTransformerBlock)
# get features
model = ViT.vit_base_patch16_224
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[[torch.Size([1, 197, 768]), torch.Size([1, 197, 768]), ...]
# change the tokens, you have to subclass ViTTokens
class MyTokens(ViTTokens):
def __init__(self, emb_size: int):
super().__init__(emb_size)
self.my_new_token = nn.Parameter(torch.randn(1, 1, emb_size))
ViT(tokens=MyTokens)
```
|
|
glasses/vit_base_patch16_384 | 2021-04-22T18:35:04.000Z | [
"pytorch",
"arxiv:2010.11929",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 12 | transformers | # vit_base_patch16_384
Implementation of Vision Transformer (ViT) proposed in [An Image Is
Worth 16x16 Words: Transformers For Image Recognition At
Scale](https://arxiv.org/pdf/2010.11929.pdf)
The following image from the authors shows the architecture.
``` python
ViT.vit_small_patch16_224()
ViT.vit_base_patch16_224()
ViT.vit_base_patch16_384()
ViT.vit_base_patch32_384()
ViT.vit_huge_patch16_224()
ViT.vit_huge_patch32_384()
ViT.vit_large_patch16_224()
ViT.vit_large_patch16_384()
ViT.vit_large_patch32_384()
```
Examples:
``` python
# change activation
ViT.vit_base_patch16_224(activation = nn.SELU)
# change number of classes (default is 1000 )
ViT.vit_base_patch16_224(n_classes=100)
# pass a different block, default is TransformerEncoderBlock
ViT.vit_base_patch16_224(block=MyCoolTransformerBlock)
# get features
model = ViT.vit_base_patch16_224
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[[torch.Size([1, 197, 768]), torch.Size([1, 197, 768]), ...]
# change the tokens, you have to subclass ViTTokens
class MyTokens(ViTTokens):
def __init__(self, emb_size: int):
super().__init__(emb_size)
self.my_new_token = nn.Parameter(torch.randn(1, 1, emb_size))
ViT(tokens=MyTokens)
```
|
|
glasses/vit_base_patch32_384 | 2021-04-22T18:37:21.000Z | [
"pytorch",
"arxiv:2010.11929",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 6 | transformers | # vit_base_patch32_384
Implementation of Vision Transformer (ViT) proposed in [An Image Is
Worth 16x16 Words: Transformers For Image Recognition At
Scale](https://arxiv.org/pdf/2010.11929.pdf)
The following image from the authors shows the architecture.
``` python
ViT.vit_small_patch16_224()
ViT.vit_base_patch16_224()
ViT.vit_base_patch16_384()
ViT.vit_base_patch32_384()
ViT.vit_huge_patch16_224()
ViT.vit_huge_patch32_384()
ViT.vit_large_patch16_224()
ViT.vit_large_patch16_384()
ViT.vit_large_patch32_384()
```
Examples:
``` python
# change activation
ViT.vit_base_patch16_224(activation = nn.SELU)
# change number of classes (default is 1000 )
ViT.vit_base_patch16_224(n_classes=100)
# pass a different block, default is TransformerEncoderBlock
ViT.vit_base_patch16_224(block=MyCoolTransformerBlock)
# get features
model = ViT.vit_base_patch16_224
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[[torch.Size([1, 197, 768]), torch.Size([1, 197, 768]), ...]
# change the tokens, you have to subclass ViTTokens
class MyTokens(ViTTokens):
def __init__(self, emb_size: int):
super().__init__(emb_size)
self.my_new_token = nn.Parameter(torch.randn(1, 1, emb_size))
ViT(tokens=MyTokens)
```
|
|
glasses/vit_huge_patch16_224 | 2021-04-22T18:39:36.000Z | [
"pytorch",
"arxiv:2010.11929",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 7 | transformers | # vit_huge_patch16_224
Implementation of Vision Transformer (ViT) proposed in [An Image Is
Worth 16x16 Words: Transformers For Image Recognition At
Scale](https://arxiv.org/pdf/2010.11929.pdf)
The following image from the authors shows the architecture.
``` python
ViT.vit_small_patch16_224()
ViT.vit_base_patch16_224()
ViT.vit_base_patch16_384()
ViT.vit_base_patch32_384()
ViT.vit_huge_patch16_224()
ViT.vit_huge_patch32_384()
ViT.vit_large_patch16_224()
ViT.vit_large_patch16_384()
ViT.vit_large_patch32_384()
```
Examples:
``` python
# change activation
ViT.vit_base_patch16_224(activation = nn.SELU)
# change number of classes (default is 1000 )
ViT.vit_base_patch16_224(n_classes=100)
# pass a different block, default is TransformerEncoderBlock
ViT.vit_base_patch16_224(block=MyCoolTransformerBlock)
# get features
model = ViT.vit_base_patch16_224
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[[torch.Size([1, 197, 768]), torch.Size([1, 197, 768]), ...]
# change the tokens, you have to subclass ViTTokens
class MyTokens(ViTTokens):
def __init__(self, emb_size: int):
super().__init__(emb_size)
self.my_new_token = nn.Parameter(torch.randn(1, 1, emb_size))
ViT(tokens=MyTokens)
```
|
|
glasses/vit_huge_patch32_384 | 2021-04-22T18:41:37.000Z | [
"pytorch",
"arxiv:2010.11929",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 11 | transformers | # vit_huge_patch32_384
Implementation of Vision Transformer (ViT) proposed in [An Image Is
Worth 16x16 Words: Transformers For Image Recognition At
Scale](https://arxiv.org/pdf/2010.11929.pdf)
The following image from the authors shows the architecture.
``` python
ViT.vit_small_patch16_224()
ViT.vit_base_patch16_224()
ViT.vit_base_patch16_384()
ViT.vit_base_patch32_384()
ViT.vit_huge_patch16_224()
ViT.vit_huge_patch32_384()
ViT.vit_large_patch16_224()
ViT.vit_large_patch16_384()
ViT.vit_large_patch32_384()
```
Examples:
``` python
# change activation
ViT.vit_base_patch16_224(activation = nn.SELU)
# change number of classes (default is 1000 )
ViT.vit_base_patch16_224(n_classes=100)
# pass a different block, default is TransformerEncoderBlock
ViT.vit_base_patch16_224(block=MyCoolTransformerBlock)
# get features
model = ViT.vit_base_patch16_224
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[[torch.Size([1, 197, 768]), torch.Size([1, 197, 768]), ...]
# change the tokens, you have to subclass ViTTokens
class MyTokens(ViTTokens):
def __init__(self, emb_size: int):
super().__init__(emb_size)
self.my_new_token = nn.Parameter(torch.randn(1, 1, emb_size))
ViT(tokens=MyTokens)
```
|
|
glasses/vit_large_patch16_224 | 2021-04-22T18:42:35.000Z | [
"pytorch",
"arxiv:2010.11929",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 7 | transformers | # vit_large_patch16_224
Implementation of Vision Transformer (ViT) proposed in [An Image Is
Worth 16x16 Words: Transformers For Image Recognition At
Scale](https://arxiv.org/pdf/2010.11929.pdf)
The following image from the authors shows the architecture.
``` python
ViT.vit_small_patch16_224()
ViT.vit_base_patch16_224()
ViT.vit_base_patch16_384()
ViT.vit_base_patch32_384()
ViT.vit_huge_patch16_224()
ViT.vit_huge_patch32_384()
ViT.vit_large_patch16_224()
ViT.vit_large_patch16_384()
ViT.vit_large_patch32_384()
```
Examples:
``` python
# change activation
ViT.vit_base_patch16_224(activation = nn.SELU)
# change number of classes (default is 1000 )
ViT.vit_base_patch16_224(n_classes=100)
# pass a different block, default is TransformerEncoderBlock
ViT.vit_base_patch16_224(block=MyCoolTransformerBlock)
# get features
model = ViT.vit_base_patch16_224
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[[torch.Size([1, 197, 768]), torch.Size([1, 197, 768]), ...]
# change the tokens, you have to subclass ViTTokens
class MyTokens(ViTTokens):
def __init__(self, emb_size: int):
super().__init__(emb_size)
self.my_new_token = nn.Parameter(torch.randn(1, 1, emb_size))
ViT(tokens=MyTokens)
```
|
|
glasses/vit_large_patch16_384 | 2021-04-22T18:43:25.000Z | [
"pytorch",
"arxiv:2010.11929",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 6 | transformers | # vit_large_patch16_384
Implementation of Vision Transformer (ViT) proposed in [An Image Is
Worth 16x16 Words: Transformers For Image Recognition At
Scale](https://arxiv.org/pdf/2010.11929.pdf)
The following image from the authors shows the architecture.
``` python
ViT.vit_small_patch16_224()
ViT.vit_base_patch16_224()
ViT.vit_base_patch16_384()
ViT.vit_base_patch32_384()
ViT.vit_huge_patch16_224()
ViT.vit_huge_patch32_384()
ViT.vit_large_patch16_224()
ViT.vit_large_patch16_384()
ViT.vit_large_patch32_384()
```
Examples:
``` python
# change activation
ViT.vit_base_patch16_224(activation = nn.SELU)
# change number of classes (default is 1000 )
ViT.vit_base_patch16_224(n_classes=100)
# pass a different block, default is TransformerEncoderBlock
ViT.vit_base_patch16_224(block=MyCoolTransformerBlock)
# get features
model = ViT.vit_base_patch16_224
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[[torch.Size([1, 197, 768]), torch.Size([1, 197, 768]), ...]
# change the tokens, you have to subclass ViTTokens
class MyTokens(ViTTokens):
def __init__(self, emb_size: int):
super().__init__(emb_size)
self.my_new_token = nn.Parameter(torch.randn(1, 1, emb_size))
ViT(tokens=MyTokens)
```
|
|
glasses/vit_large_patch32_384 | 2021-04-22T18:44:12.000Z | [
"pytorch",
"arxiv:2010.11929",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 6 | transformers | # vit_large_patch32_384
Implementation of Vision Transformer (ViT) proposed in [An Image Is
Worth 16x16 Words: Transformers For Image Recognition At
Scale](https://arxiv.org/pdf/2010.11929.pdf)
The following image from the authors shows the architecture.
``` python
ViT.vit_small_patch16_224()
ViT.vit_base_patch16_224()
ViT.vit_base_patch16_384()
ViT.vit_base_patch32_384()
ViT.vit_huge_patch16_224()
ViT.vit_huge_patch32_384()
ViT.vit_large_patch16_224()
ViT.vit_large_patch16_384()
ViT.vit_large_patch32_384()
```
Examples:
``` python
# change activation
ViT.vit_base_patch16_224(activation = nn.SELU)
# change number of classes (default is 1000 )
ViT.vit_base_patch16_224(n_classes=100)
# pass a different block, default is TransformerEncoderBlock
ViT.vit_base_patch16_224(block=MyCoolTransformerBlock)
# get features
model = ViT.vit_base_patch16_224
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[[torch.Size([1, 197, 768]), torch.Size([1, 197, 768]), ...]
# change the tokens, you have to subclass ViTTokens
class MyTokens(ViTTokens):
def __init__(self, emb_size: int):
super().__init__(emb_size)
self.my_new_token = nn.Parameter(torch.randn(1, 1, emb_size))
ViT(tokens=MyTokens)
```
|
|
glasses/wide_resnet101_2 | 2021-04-24T10:52:37.000Z | [
"pytorch",
"arxiv:1605.07146",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 9 | transformers | # wide_resnet101_2
Implementation of Wide ResNet proposed in [\"Wide Residual
Networks\"](https://arxiv.org/pdf/1605.07146.pdf)
Create a default model
``` python
WideResNet.wide_resnet50_2()
WideResNet.wide_resnet101_2()
# create a wide_resnet18_4
WideResNet.resnet18(block=WideResNetBottleNeckBlock, width_factor=4)
```
Examples:
``` python
# change activation
WideResNet.resnext50_32x4d(activation = nn.SELU)
# change number of classes (default is 1000 )
WideResNet.resnext50_32x4d(n_classes=100)
# pass a different block
WideResNet.resnext50_32x4d(block=SENetBasicBlock)
# change the initial convolution
model = WideResNet.resnext50_32x4d
model.encoder.gate.conv1 = nn.Conv2d(3, 64, kernel_size=3)
# store each feature
x = torch.rand((1, 3, 224, 224))
model = WideResNet.wide_resnet50_2()
features = []
x = model.encoder.gate(x)
for block in model.encoder.layers:
x = block(x)
features.append(x)
print([x.shape for x in features])
# [torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14]), torch.Size([1, 512, 7, 7])]
```
|
|
glasses/wide_resnet50_2 | 2021-04-24T10:52:04.000Z | [
"pytorch",
"arxiv:1605.07146",
"transformers"
]
| [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin"
]
| glasses | 6 | transformers | # wide_resnet50_2
Implementation of Wide ResNet proposed in [\"Wide Residual
Networks\"](https://arxiv.org/pdf/1605.07146.pdf)
Create a default model
``` python
WideResNet.wide_resnet50_2()
WideResNet.wide_resnet101_2()
# create a wide_resnet18_4
WideResNet.resnet18(block=WideResNetBottleNeckBlock, width_factor=4)
```
Examples:
``` python
# change activation
WideResNet.resnext50_32x4d(activation = nn.SELU)
# change number of classes (default is 1000 )
WideResNet.resnext50_32x4d(n_classes=100)
# pass a different block
WideResNet.resnext50_32x4d(block=SENetBasicBlock)
# change the initial convolution
model = WideResNet.resnext50_32x4d
model.encoder.gate.conv1 = nn.Conv2d(3, 64, kernel_size=3)
# store each feature
x = torch.rand((1, 3, 224, 224))
model = WideResNet.wide_resnet50_2()
features = []
x = model.encoder.gate(x)
for block in model.encoder.layers:
x = block(x)
features.append(x)
print([x.shape for x in features])
# [torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14]), torch.Size([1, 512, 7, 7])]
```
|
|
gmihaila/distilbert-base-uncased | 2020-11-05T02:26:13.000Z | [
"pytorch",
"distilbert",
"text-classification",
"transformers"
]
| text-classification | [
".gitattributes",
"config.json",
"pytorch_model.bin",
"special_tokens_map.json",
"tokenizer_config.json",
"vocab.txt"
]
| gmihaila | 15 | transformers | |
gmihaila/wav2vec2-large-xlsr-53-romanian | 2021-03-23T06:15:19.000Z | [
"pytorch",
"wav2vec2",
"ro",
"dataset:common_voice",
"transformers",
"audio",
"automatic-speech-recognition",
"speech",
"xlsr-fine-tuning-week",
"license:apache-2.0"
]
| automatic-speech-recognition | [
".gitattributes",
"README.md",
"config.json",
"preprocessor_config.json",
"pytorch_model.bin",
"special_tokens_map.json",
"tokenizer_config.json",
"vocab.json"
]
| gmihaila | 121 | transformers | ---
language: ro
datasets:
- common_voice
tags:
- audio
- automatic-speech-recognition
- speech
- xlsr-fine-tuning-week
license: apache-2.0
model-index:
- name: XLSR Wav2Vec2 Romanian by George Mihaila
results:
- task:
name: Speech Recognition
type: automatic-speech-recognition
dataset:
name: Common Voice ro
type: common_voice
args: ro
metrics:
- name: Test WER
type: wer
value: 28.4
---
# Wav2Vec2-Large-XLSR-53-Romanian
Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) in Romanian using the [Common Voice](https://huggingface.co/datasets/common_voice)
When using this model, make sure that your speech input is sampled at 16kHz.
## Usage
The model can be used directly (without a language model) as follows:
```python
import torch
import torchaudio
from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
test_dataset = load_dataset("common_voice", "ro", split="test[:2%]").
processor = Wav2Vec2Processor.from_pretrained("gmihaila/wav2vec2-large-xlsr-53-romanian")
model = Wav2Vec2ForCTC.from_pretrained("gmihaila/wav2vec2-large-xlsr-53-romanian")
resampler = torchaudio.transforms.Resample(48_000, 16_000)
# Preprocessing the datasets.
# We need to read the aduio files as arrays
def speech_file_to_array_fn(batch):
\\\\tspeech_array, sampling_rate = torchaudio.load(batch["path"])
\\\\tbatch["speech"] = resampler(speech_array).squeeze().numpy()
\\\\treturn batch
test_dataset = test_dataset.map(speech_file_to_array_fn)
inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
with torch.no_grad():
\\\\tlogits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
predicted_ids = torch.argmax(logits, dim=-1)
print("Prediction:", processor.batch_decode(predicted_ids))
print("Reference:", test_dataset["sentence"][:2])
```
## Evaluation
The model can be evaluated as follows on the {language} test data of Common Voice.
```python
import torch
import torchaudio
from datasets import load_dataset, load_metric
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import re
test_dataset = load_dataset("common_voice", "ro", split="test")
wer = load_metric("wer")
processor = Wav2Vec2Processor.from_pretrained("gmihaila/wav2vec2-large-xlsr-53-romanian")
model = Wav2Vec2ForCTC.from_pretrained("gmihaila/wav2vec2-large-xlsr-53-romanian")
model.to("cuda")
chars_to_ignore_regex = '[\\\\\\\\,\\\\\\\\?\\\\\\\\.\\\\\\\\!\\\\\\\\-\\\\\\\\;\\\\\\\\:\\\\\\\\"\\\\\\\\“]'
resampler = torchaudio.transforms.Resample(48_000, 16_000)
# Preprocessing the datasets.
# We need to read the aduio files as arrays
def speech_file_to_array_fn(batch):
\\\\tbatch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
\\\\tspeech_array, sampling_rate = torchaudio.load(batch["path"])
\\\\tbatch["speech"] = resampler(speech_array).squeeze().numpy()
\\\\treturn batch
test_dataset = test_dataset.map(speech_file_to_array_fn)
# Preprocessing the datasets.
# We need to read the aduio files as arrays
def evaluate(batch):
\\\\tinputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
\\\\twith torch.no_grad():
\\\\t\\\\tlogits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
pred_ids = torch.argmax(logits, dim=-1)
\\\\tbatch["pred_strings"] = processor.batch_decode(pred_ids)
\\\\treturn batch
result = test_dataset.map(evaluate, batched=True, batch_size=8)
print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
```
**Test Result**: 28.43 %
## Training
The Common Voice `train`, `validation` datasets were used for training.
The script used for training can be found [here](https://colab.research.google.com/github/gmihaila/ml_things/blob/master/notebooks/pytorch/RO_Fine_Tune_XLSR_Wav2Vec2_on_Turkish_ASR_with_🤗_Transformers.ipynb) |
gogamza/kobart-summarization | 2021-05-20T06:35:05.000Z | [
"pytorch",
"bart",
"seq2seq",
"ko",
"transformers",
"license:mit",
"text2text-generation"
]
| text2text-generation | [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin",
"summ.png",
"tokenizer.json"
]
| gogamza | 193 | transformers | ---
language: ko
tags:
- bart
license: MIT
---
# Korean News Summarization Model
## How to use
```python
import torch
from transformers import PreTrainedTokenizerFast
from transformers import BartForConditionalGeneration
tokenizer = PreTrainedTokenizerFast.from_pretrained(
'gogamza/kobart-summarization',
bos_token='<s>', eos_token='</s>', unk_token='<unk>', pad_token='<pad>', mask_token='<mask>')
model = BartForConditionalGeneration.from_pretrained('gogamza/kobart-summarization')
text = "과거를 떠올려보자. 방송을 보던 우리의 모습을..."
raw_input_ids = tokenizer.encode(text)
input_ids = [tokenizer.bos_token_id] + \\
raw_input_ids + [tokenizer.eos_token_id]
summary_ids = model.generate(torch.tensor([input_ids]),
max_length=150,
early_stopping=False,
num_beams=5,
repetition_penalty=1.0,
eos_token_id=tokenizer.eos_token_id)
summ_text = tokenizer.batch_decode(summary_ids.tolist(), skip_special_tokens=True)[0]
```
|
gogamza/kogpt2 | 2021-04-30T00:08:22.000Z | []
| [
".gitattributes"
]
| gogamza | 0 | |||
gonced8/pegasus-qa | 2021-05-17T15:46:42.000Z | [
"pytorch",
"pegasus",
"seq2seq",
"transformers",
"text2text-generation"
]
| text2text-generation | [
".gitattributes",
"config.json",
"pytorch_model.bin",
"special_tokens_map.json",
"spiece.model",
"tokenizer_config.json"
]
| gonced8 | 28 | transformers | |
google/bert2bert_L-24_wmt_de_en | 2020-12-11T21:41:14.000Z | [
"pytorch",
"encoder-decoder",
"seq2seq",
"en",
"de",
"dataset:wmt14",
"arxiv:1907.12461",
"transformers",
"license:apache-2.0",
"translation",
"text2text-generation"
]
| translation | [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin",
"vocab.txt"
]
| google | 264 | transformers | ---
language:
- en
- de
license: apache-2.0
datasets:
- wmt14
tags:
- translation
---
# bert2bert_L-24_wmt_de_en EncoderDecoder model
The model was introduced in
[this paper](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn and first released in [this repository](https://tfhub.dev/google/bertseq2seq/bert24_de_en/1).
The model is an encoder-decoder model that was initialized on the `bert-large` checkpoints for both the encoder
and decoder and fine-tuned on German to English translation on the WMT dataset, which is linked above.
Disclaimer: The model card has been written by the Hugging Face team.
## How to use
You can use this model for translation, *e.g.*
```python
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
tokenizer = AutoTokenizer.from_pretrained("google/bert2bert_L-24_wmt_de_en", pad_token="<pad>", eos_token="</s>", bos_token="<s>")
model = AutoModelForSeq2SeqLM.from_pretrained("google/bert2bert_L-24_wmt_de_en")
sentence = "Willst du einen Kaffee trinken gehen mit mir?"
input_ids = tokenizer(sentence, return_tensors="pt", add_special_tokens=False).input_ids
output_ids = model.generate(input_ids)[0]
print(tokenizer.decode(output_ids, skip_special_tokens=True))
# should output
# Want to drink a kaffee go with me? .
```
|
google/bert2bert_L-24_wmt_en_de | 2020-12-11T21:41:17.000Z | [
"pytorch",
"encoder-decoder",
"seq2seq",
"en",
"de",
"dataset:wmt14",
"arxiv:1907.12461",
"transformers",
"license:apache-2.0",
"translation",
"text2text-generation"
]
| translation | [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin",
"vocab.txt"
]
| google | 96 | transformers | ---
language:
- en
- de
license: apache-2.0
datasets:
- wmt14
tags:
- translation
---
# bert2bert_L-24_wmt_en_de EncoderDecoder model
The model was introduced in
[this paper](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn and first released in [this repository](https://tfhub.dev/google/bertseq2seq/bert24_en_de/1).
The model is an encoder-decoder model that was initialized on the `bert-large` checkpoints for both the encoder
and decoder and fine-tuned on English to German translation on the WMT dataset, which is linked above.
Disclaimer: The model card has been written by the Hugging Face team.
## How to use
You can use this model for translation, *e.g.*
```python
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
tokenizer = AutoTokenizer.from_pretrained("google/bert2bert_L-24_wmt_en_de", pad_token="<pad>", eos_token="</s>", bos_token="<s>")
model = AutoModelForSeq2SeqLM.from_pretrained("google/bert2bert_L-24_wmt_en_de")
sentence = "Would you like to grab a coffee with me this week?"
input_ids = tokenizer(sentence, return_tensors="pt", add_special_tokens=False).input_ids
output_ids = model.generate(input_ids)[0]
print(tokenizer.decode(output_ids, skip_special_tokens=True))
# should output
# Möchten Sie diese Woche einen Kaffee mit mir schnappen?
|
google/bert_for_seq_generation_L-24_bbc_encoder | 2020-09-11T07:57:22.000Z | [
"pytorch",
"bert-generation",
"transformers"
]
| [
".gitattributes",
"config.json",
"pytorch_model.bin",
"spiece.model"
]
| google | 616 | transformers | ||
google/bert_uncased_L-10_H-128_A-2 | 2021-05-19T17:23:15.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 31 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-10_H-256_A-4 | 2021-05-19T17:23:44.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 116 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-10_H-512_A-8 | 2021-05-19T17:24:16.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 33 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-10_H-768_A-12 | 2021-05-19T17:24:59.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 19 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-12_H-128_A-2 | 2021-05-19T17:26:01.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 1,582 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-12_H-256_A-4 | 2021-05-19T17:26:24.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 544 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-12_H-512_A-8 | 2021-05-19T17:26:55.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 846,108 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-12_H-768_A-12 | 2021-05-19T17:27:43.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 3,149 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-2_H-128_A-2 | 2021-05-19T17:28:12.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 55,317 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-2_H-256_A-4 | 2021-05-19T17:28:46.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 181 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-2_H-512_A-8 | 2021-05-19T17:29:08.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 161 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-2_H-768_A-12 | 2021-05-19T17:29:34.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 48 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-4_H-128_A-2 | 2021-05-19T17:30:08.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 40 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-4_H-256_A-4 | 2021-05-19T17:30:27.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 14,941 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-4_H-512_A-8 | 2021-05-19T17:30:51.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 13,808 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-4_H-768_A-12 | 2021-05-19T17:31:28.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 38 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-6_H-128_A-2 | 2021-05-19T17:33:17.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 21 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-6_H-256_A-4 | 2021-05-19T17:33:36.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 20 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-6_H-512_A-8 | 2021-05-19T17:34:01.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 98 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-6_H-768_A-12 | 2021-05-19T17:34:36.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 637 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-8_H-128_A-2 | 2021-05-19T17:35:05.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 18 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-8_H-256_A-4 | 2021-05-19T17:35:25.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 35 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-8_H-512_A-8 | 2021-05-19T17:35:53.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 10,748 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bert_uncased_L-8_H-768_A-12 | 2021-05-19T17:36:32.000Z | [
"pytorch",
"jax",
"bert",
"arxiv:1908.08962",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"bert_model.ckpt.data-00000-of-00001",
"bert_model.ckpt.index",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"vocab.txt"
]
| google | 230 | transformers | ---
thumbnail: https://huggingface.co/front/thumbnails/google.png
license: apache-2.0
---
BERT Miniatures
===
This is the set of 24 BERT models referenced in [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://arxiv.org/abs/1908.08962) (English only, uncased, trained with WordPiece masking).
We have shown that the standard BERT recipe (including model architecture and training objective) is effective on a wide range of model sizes, beyond BERT-Base and BERT-Large. The smaller BERT models are intended for environments with restricted computational resources. They can be fine-tuned in the same manner as the original BERT models. However, they are most effective in the context of knowledge distillation, where the fine-tuning labels are produced by a larger and more accurate teacher.
Our goal is to enable research in institutions with fewer computational resources and encourage the community to seek directions of innovation alternative to increasing model capacity.
You can download the 24 BERT miniatures either from the [official BERT Github page](https://github.com/google-research/bert/), or via HuggingFace from the links below:
| |H=128|H=256|H=512|H=768|
|---|:---:|:---:|:---:|:---:|
| **L=2** |[**2/128 (BERT-Tiny)**][2_128]|[2/256][2_256]|[2/512][2_512]|[2/768][2_768]|
| **L=4** |[4/128][4_128]|[**4/256 (BERT-Mini)**][4_256]|[**4/512 (BERT-Small)**][4_512]|[4/768][4_768]|
| **L=6** |[6/128][6_128]|[6/256][6_256]|[6/512][6_512]|[6/768][6_768]|
| **L=8** |[8/128][8_128]|[8/256][8_256]|[**8/512 (BERT-Medium)**][8_512]|[8/768][8_768]|
| **L=10** |[10/128][10_128]|[10/256][10_256]|[10/512][10_512]|[10/768][10_768]|
| **L=12** |[12/128][12_128]|[12/256][12_256]|[12/512][12_512]|[**12/768 (BERT-Base)**][12_768]|
Note that the BERT-Base model in this release is included for completeness only; it was re-trained under the same regime as the original model.
Here are the corresponding GLUE scores on the test set:
|Model|Score|CoLA|SST-2|MRPC|STS-B|QQP|MNLI-m|MNLI-mm|QNLI(v2)|RTE|WNLI|AX|
|---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|BERT-Tiny|64.2|0.0|83.2|81.1/71.1|74.3/73.6|62.2/83.4|70.2|70.3|81.5|57.2|62.3|21.0|
|BERT-Mini|65.8|0.0|85.9|81.1/71.8|75.4/73.3|66.4/86.2|74.8|74.3|84.1|57.9|62.3|26.1|
|BERT-Small|71.2|27.8|89.7|83.4/76.2|78.8/77.0|68.1/87.0|77.6|77.0|86.4|61.8|62.3|28.6|
|BERT-Medium|73.5|38.0|89.6|86.6/81.6|80.4/78.4|69.6/87.9|80.0|79.1|87.7|62.2|62.3|30.5|
For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained for 4 epochs:
- batch sizes: 8, 16, 32, 64, 128
- learning rates: 3e-4, 1e-4, 5e-5, 3e-5
If you use these models, please cite the following paper:
```
@article{turc2019,
title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
journal={arXiv preprint arXiv:1908.08962v2 },
year={2019}
}
```
[2_128]: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
[2_256]: https://huggingface.co/google/bert_uncased_L-2_H-256_A-4
[2_512]: https://huggingface.co/google/bert_uncased_L-2_H-512_A-8
[2_768]: https://huggingface.co/google/bert_uncased_L-2_H-768_A-12
[4_128]: https://huggingface.co/google/bert_uncased_L-4_H-128_A-2
[4_256]: https://huggingface.co/google/bert_uncased_L-4_H-256_A-4
[4_512]: https://huggingface.co/google/bert_uncased_L-4_H-512_A-8
[4_768]: https://huggingface.co/google/bert_uncased_L-4_H-768_A-12
[6_128]: https://huggingface.co/google/bert_uncased_L-6_H-128_A-2
[6_256]: https://huggingface.co/google/bert_uncased_L-6_H-256_A-4
[6_512]: https://huggingface.co/google/bert_uncased_L-6_H-512_A-8
[6_768]: https://huggingface.co/google/bert_uncased_L-6_H-768_A-12
[8_128]: https://huggingface.co/google/bert_uncased_L-8_H-128_A-2
[8_256]: https://huggingface.co/google/bert_uncased_L-8_H-256_A-4
[8_512]: https://huggingface.co/google/bert_uncased_L-8_H-512_A-8
[8_768]: https://huggingface.co/google/bert_uncased_L-8_H-768_A-12
[10_128]: https://huggingface.co/google/bert_uncased_L-10_H-128_A-2
[10_256]: https://huggingface.co/google/bert_uncased_L-10_H-256_A-4
[10_512]: https://huggingface.co/google/bert_uncased_L-10_H-512_A-8
[10_768]: https://huggingface.co/google/bert_uncased_L-10_H-768_A-12
[12_128]: https://huggingface.co/google/bert_uncased_L-12_H-128_A-2
[12_256]: https://huggingface.co/google/bert_uncased_L-12_H-256_A-4
[12_512]: https://huggingface.co/google/bert_uncased_L-12_H-512_A-8
[12_768]: https://huggingface.co/google/bert_uncased_L-12_H-768_A-12
|
|
google/bigbird-base-trivia-itc | 2021-06-02T14:53:34.000Z | [
"pytorch",
"jax",
"big_bird",
"question-answering",
"en",
"dataset:trivia_qa",
"arxiv:2007.14062",
"transformers",
"license:apache-2.0"
]
| question-answering | [
".gitattributes",
"README.md",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"special_tokens_map.json",
"spiece.model",
"tokenizer_config.json"
]
| google | 3,471 | transformers | ---
language: en
license: apache-2.0
datasets:
- trivia_qa
---
# BigBird base trivia-itc
This model is a fine-tune checkpoint of `bigbird-roberta-base`, fine-tuned on `trivia_qa` with `BigBirdForQuestionAnsweringHead` on its top.
Check out [this](https://colab.research.google.com/drive/1DVOm1VHjW0eKCayFq1N2GpY6GR9M4tJP?usp=sharing) to see how well `google/bigbird-base-trivia-itc` performs on question answering.
## How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BigBirdForQuestionAnswering
# by default its in `block_sparse` mode with num_random_blocks=3, block_size=64
model = BigBirdForQuestionAnswering.from_pretrained("google/bigbird-base-trivia-itc")
# you can change `attention_type` to full attention like this:
model = BigBirdForQuestionAnswering.from_pretrained("google/bigbird-base-trivia-itc", attention_type="original_full")
# you can change `block_size` & `num_random_blocks` like this:
model = BigBirdForQuestionAnswering.from_pretrained("google/bigbird-base-trivia-itc", block_size=16, num_random_blocks=2)
question = "Replace me by any text you'd like."
context = "Put some context for answering"
encoded_input = tokenizer(question, context, return_tensors='pt')
output = model(**encoded_input)
```
# Fine-tuning config & hyper-parameters
- No. of global token = 128
- Window length = 192
- No. of random token = 192
- Max. sequence length = 4096
- No. of heads = 12
- No. of hidden layers = 12
- Hidden layer size = 768
- Batch size = 32
- Loss = cross-entropy noisy spans
## BibTeX entry and citation info
```tex
@misc{zaheer2021big,
title={Big Bird: Transformers for Longer Sequences},
author={Manzil Zaheer and Guru Guruganesh and Avinava Dubey and Joshua Ainslie and Chris Alberti and Santiago Ontanon and Philip Pham and Anirudh Ravula and Qifan Wang and Li Yang and Amr Ahmed},
year={2021},
eprint={2007.14062},
archivePrefix={arXiv},
primaryClass={cs.LG}
}
```
|
google/bigbird-pegasus-large-arxiv | 2021-06-03T18:26:02.000Z | [
"pytorch",
"bigbird_pegasus",
"seq2seq",
"en",
"dataset:scientific_papers",
"arxiv:2007.14062",
"transformers",
"license:apache-2.0",
"summarization",
"text2text-generation"
]
| summarization | [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin",
"special_tokens_map.json",
"spiece.model",
"tokenizer.json",
"tokenizer_config.json"
]
| google | 1,596 | transformers | ---
language: en
license: apache-2.0
datasets:
- scientific_papers
tags:
- summarization
---
# BigBirdPegasus model (large)
BigBird, is a sparse-attention based transformer which extends Transformer based models, such as BERT to much longer sequences. Moreover, BigBird comes along with a theoretical understanding of the capabilities of a complete transformer that the sparse model can handle.
BigBird was introduced in this [paper](https://arxiv.org/abs/2007.14062) and first released in this [repository](https://github.com/google-research/bigbird).
Disclaimer: The team releasing BigBird did not write a model card for this model so this model card has been written by the Hugging Face team.
## Model description
BigBird relies on **block sparse attention** instead of normal attention (i.e. BERT's attention) and can handle sequences up to a length of 4096 at a much lower compute cost compared to BERT. It has achieved SOTA on various tasks involving very long sequences such as long documents summarization, question-answering with long contexts.
## How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BigBirdPegasusForConditionalGeneration, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("google/bigbird-pegasus-large-arxiv")
# by default encoder-attention is `block_sparse` with num_random_blocks=3, block_size=64
model = BigBirdPegasusForConditionalGeneration.from_pretrained("google/bigbird-pegasus-large-arxiv")
# decoder attention type can't be changed & will be "original_full"
# you can change `attention_type` (encoder only) to full attention like this:
model = BigBirdPegasusForConditionalGeneration.from_pretrained("google/bigbird-pegasus-large-arxiv", attention_type="original_full")
# you can change `block_size` & `num_random_blocks` like this:
model = BigBirdPegasusForConditionalGeneration.from_pretrained("google/bigbird-pegasus-large-arxiv", block_size=16, num_random_blocks=2)
text = "Replace me by any text you'd like."
inputs = tokenizer(text, return_tensors='pt')
prediction = model.generate(**inputs)
prediction = tokenizer.batch_decode(prediction)
```
## Training Procedure
This checkpoint is obtained after fine-tuning `BigBirdPegasusForConditionalGeneration` for **summarization** on **arxiv dataset** from [scientific_papers](https://huggingface.co/datasets/scientific_papers).
## BibTeX entry and citation info
```tex
@misc{zaheer2021big,
title={Big Bird: Transformers for Longer Sequences},
author={Manzil Zaheer and Guru Guruganesh and Avinava Dubey and Joshua Ainslie and Chris Alberti and Santiago Ontanon and Philip Pham and Anirudh Ravula and Qifan Wang and Li Yang and Amr Ahmed},
year={2021},
eprint={2007.14062},
archivePrefix={arXiv},
primaryClass={cs.LG}
}
```
|
google/bigbird-pegasus-large-bigpatent | 2021-06-03T18:26:21.000Z | [
"pytorch",
"bigbird_pegasus",
"seq2seq",
"en",
"dataset:big_patent",
"arxiv:2007.14062",
"transformers",
"license:apache-2.0",
"summarization",
"text2text-generation"
]
| summarization | [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin",
"special_tokens_map.json",
"spiece.model",
"tokenizer.json",
"tokenizer_config.json"
]
| google | 1,012 | transformers | ---
language: en
license: apache-2.0
datasets:
- big_patent
tags:
- summarization
---
# BigBirdPegasus model (large)
BigBird, is a sparse-attention based transformer which extends Transformer based models, such as BERT to much longer sequences. Moreover, BigBird comes along with a theoretical understanding of the capabilities of a complete transformer that the sparse model can handle.
BigBird was introduced in this [paper](https://arxiv.org/abs/2007.14062) and first released in this [repository](https://github.com/google-research/bigbird).
Disclaimer: The team releasing BigBird did not write a model card for this model so this model card has been written by the Hugging Face team.
## Model description
BigBird relies on **block sparse attention** instead of normal attention (i.e. BERT's attention) and can handle sequences up to a length of 4096 at a much lower compute cost compared to BERT. It has achieved SOTA on various tasks involving very long sequences such as long documents summarization, question-answering with long contexts.
## How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BigBirdPegasusForConditionalGeneration, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("google/bigbird-pegasus-large-bigpatent")
# by default encoder-attention is `block_sparse` with num_random_blocks=3, block_size=64
model = BigBirdPegasusForConditionalGeneration.from_pretrained("google/bigbird-pegasus-large-bigpatent")
# decoder attention type can't be changed & will be "original_full"
# you can change `attention_type` (encoder only) to full attention like this:
model = BigBirdPegasusForConditionalGeneration.from_pretrained("google/bigbird-pegasus-large-bigpatent", attention_type="original_full")
# you can change `block_size` & `num_random_blocks` like this:
model = BigBirdPegasusForConditionalGeneration.from_pretrained("google/bigbird-pegasus-large-bigpatent", block_size=16, num_random_blocks=2)
text = "Replace me by any text you'd like."
inputs = tokenizer(text, return_tensors='pt')
prediction = model.generate(**inputs)
prediction = tokenizer.batch_decode(prediction)
```
## Training Procedure
This checkpoint is obtained after fine-tuning `BigBirdPegasusForConditionalGeneration` for **summarization** on [big_patent](https://huggingface.co/datasets/big_patent) dataset.
## BibTeX entry and citation info
```tex
@misc{zaheer2021big,
title={Big Bird: Transformers for Longer Sequences},
author={Manzil Zaheer and Guru Guruganesh and Avinava Dubey and Joshua Ainslie and Chris Alberti and Santiago Ontanon and Philip Pham and Anirudh Ravula and Qifan Wang and Li Yang and Amr Ahmed},
year={2021},
eprint={2007.14062},
archivePrefix={arXiv},
primaryClass={cs.LG}
}
```
|
google/bigbird-pegasus-large-pubmed | 2021-06-03T18:25:30.000Z | [
"pytorch",
"bigbird_pegasus",
"seq2seq",
"en",
"dataset:scientific_papers",
"arxiv:2007.14062",
"transformers",
"license:apache-2.0",
"summarization",
"text2text-generation"
]
| summarization | [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin",
"special_tokens_map.json",
"spiece.model",
"tokenizer.json",
"tokenizer_config.json"
]
| google | 1,554 | transformers | ---
language: en
license: apache-2.0
datasets:
- scientific_papers
tags:
- summarization
---
# BigBirdPegasus model (large)
BigBird, is a sparse-attention based transformer which extends Transformer based models, such as BERT to much longer sequences. Moreover, BigBird comes along with a theoretical understanding of the capabilities of a complete transformer that the sparse model can handle.
BigBird was introduced in this [paper](https://arxiv.org/abs/2007.14062) and first released in this [repository](https://github.com/google-research/bigbird).
Disclaimer: The team releasing BigBird did not write a model card for this model so this model card has been written by the Hugging Face team.
## Model description
BigBird relies on **block sparse attention** instead of normal attention (i.e. BERT's attention) and can handle sequences up to a length of 4096 at a much lower compute cost compared to BERT. It has achieved SOTA on various tasks involving very long sequences such as long documents summarization, question-answering with long contexts.
## How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BigBirdPegasusForConditionalGeneration, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("google/bigbird-pegasus-large-pubmed")
# by default encoder-attention is `block_sparse` with num_random_blocks=3, block_size=64
model = BigBirdPegasusForConditionalGeneration.from_pretrained("google/bigbird-pegasus-large-pubmed")
# decoder attention type can't be changed & will be "original_full"
# you can change `attention_type` (encoder only) to full attention like this:
model = BigBirdPegasusForConditionalGeneration.from_pretrained("google/bigbird-pegasus-large-pubmed", attention_type="original_full")
# you can change `block_size` & `num_random_blocks` like this:
model = BigBirdPegasusForConditionalGeneration.from_pretrained("google/bigbird-pegasus-large-pubmed", block_size=16, num_random_blocks=2)
text = "Replace me by any text you'd like."
inputs = tokenizer(text, return_tensors='pt')
prediction = model.generate(**inputs)
prediction = tokenizer.batch_decode(prediction)
```
## Training Procedure
This checkpoint is obtained after fine-tuning `BigBirdPegasusForConditionalGeneration` for **summarization** on **pubmed dataset** from [scientific_papers](https://huggingface.co/datasets/scientific_papers).
## BibTeX entry and citation info
```tex
@misc{zaheer2021big,
title={Big Bird: Transformers for Longer Sequences},
author={Manzil Zaheer and Guru Guruganesh and Avinava Dubey and Joshua Ainslie and Chris Alberti and Santiago Ontanon and Philip Pham and Anirudh Ravula and Qifan Wang and Li Yang and Amr Ahmed},
year={2021},
eprint={2007.14062},
archivePrefix={arXiv},
primaryClass={cs.LG}
}
```
|
google/bigbird-roberta-base | 2021-06-02T14:30:54.000Z | [
"pytorch",
"jax",
"big_bird",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"dataset:cc_news",
"arxiv:2007.14062",
"transformers",
"license:apache-2.0"
]
| [
".gitattributes",
"README.md",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"special_tokens_map.json",
"spiece.model",
"tokenizer_config.json"
]
| google | 66,150 | transformers | ---
language: en
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
- cc_news
---
# BigBird base model
BigBird, is a sparse-attention based transformer which extends Transformer based models, such as BERT to much longer sequences. Moreover, BigBird comes along with a theoretical understanding of the capabilities of a complete transformer that the sparse model can handle.
It is a pretrained model on English language using a masked language modeling (MLM) objective. It was introduced in this [paper](https://arxiv.org/abs/2007.14062) and first released in this [repository](https://github.com/google-research/bigbird).
Disclaimer: The team releasing BigBird did not write a model card for this model so this model card has been written by the Hugging Face team.
## Model description
BigBird relies on **block sparse attention** instead of normal attention (i.e. BERT's attention) and can handle sequences up to a length of 4096 at a much lower compute cost compared to BERT. It has achieved SOTA on various tasks involving very long sequences such as long documents summarization, question-answering with long contexts.
## How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BigBirdModel
# by default its in `block_sparse` mode with num_random_blocks=3, block_size=64
model = BigBirdModel.from_pretrained("google/bigbird-roberta-base")
# you can change `attention_type` to full attention like this:
model = BigBirdModel.from_pretrained("google/bigbird-roberta-base", attention_type="original_full")
# you can change `block_size` & `num_random_blocks` like this:
model = BigBirdModel.from_pretrained("google/bigbird-roberta-base", block_size=16, num_random_blocks=2)
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
## Training Data
This model is pre-trained on four publicly available datasets: **Books**, **CC-News**, **Stories** and **Wikipedia**. It used same sentencepiece vocabulary as RoBERTa (which is in turn borrowed from GPT2).
## Training Procedure
Document longer than 4096 were split into multiple documents and documents that were much smaller than 4096 were joined. Following the original BERT training, 15% of tokens were masked and model is trained to predict the mask.
Model is warm started from RoBERTa’s checkpoint.
## BibTeX entry and citation info
```tex
@misc{zaheer2021big,
title={Big Bird: Transformers for Longer Sequences},
author={Manzil Zaheer and Guru Guruganesh and Avinava Dubey and Joshua Ainslie and Chris Alberti and Santiago Ontanon and Philip Pham and Anirudh Ravula and Qifan Wang and Li Yang and Amr Ahmed},
year={2021},
eprint={2007.14062},
archivePrefix={arXiv},
primaryClass={cs.LG}
}
```
|
|
google/bigbird-roberta-large | 2021-06-02T14:49:29.000Z | [
"pytorch",
"jax",
"big_bird",
"masked-lm",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"dataset:cc_news",
"arxiv:2007.14062",
"transformers",
"license:apache-2.0",
"fill-mask"
]
| fill-mask | [
".gitattributes",
"README.md",
"config.json",
"flax_model.msgpack",
"pytorch_model.bin",
"special_tokens_map.json",
"spiece.model",
"tokenizer_config.json"
]
| google | 1,138 | transformers | ---
language: en
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
- cc_news
---
# BigBird large model
BigBird, is a sparse-attention based transformer which extends Transformer based models, such as BERT to much longer sequences. Moreover, BigBird comes along with a theoretical understanding of the capabilities of a complete transformer that the sparse model can handle.
It is a pretrained model on English language using a masked language modeling (MLM) objective. It was introduced in this [paper](https://arxiv.org/abs/2007.14062) and first released in this [repository](https://github.com/google-research/bigbird).
Disclaimer: The team releasing BigBird did not write a model card for this model so this model card has been written by the Hugging Face team.
## Model description
BigBird relies on **block sparse attention** instead of normal attention (i.e. BERT's attention) and can handle sequences up to a length of 4096 at a much lower compute cost compared to BERT. It has achieved SOTA on various tasks involving very long sequences such as long documents summarization, question-answering with long contexts.
## How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BigBirdModel
# by default its in `block_sparse` mode with num_random_blocks=3, block_size=64
model = BigBirdModel.from_pretrained("google/bigbird-roberta-large")
# you can change `attention_type` to full attention like this:
model = BigBirdModel.from_pretrained("google/bigbird-roberta-large", attention_type="original_full")
# you can change `block_size` & `num_random_blocks` like this:
model = BigBirdModel.from_pretrained("google/bigbird-roberta-large", block_size=16, num_random_blocks=2)
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
## Training Data
This model is pre-trained on four publicly available datasets: **Books**, **CC-News**, **Stories** and **Wikipedia**. It used same sentencepiece vocabulary as RoBERTa (which is in turn borrowed from GPT2).
## Training Procedure
Document longer than 4096 were split into multiple documents and documents that were much smaller than 4096 were joined. Following the original BERT training, 15% of tokens were masked and model is trained to predict the mask.
Model is warm started from RoBERTa’s checkpoint.
## BibTeX entry and citation info
```tex
@misc{zaheer2021big,
title={Big Bird: Transformers for Longer Sequences},
author={Manzil Zaheer and Guru Guruganesh and Avinava Dubey and Joshua Ainslie and Chris Alberti and Santiago Ontanon and Philip Pham and Anirudh Ravula and Qifan Wang and Li Yang and Amr Ahmed},
year={2021},
eprint={2007.14062},
archivePrefix={arXiv},
primaryClass={cs.LG}
}
```
|
google/byt5-base | 2021-06-01T17:48:30.000Z | [
"pytorch",
"tf",
"t5",
"seq2seq",
"multilingual",
"dataset:mc4",
"arxiv:1907.06292",
"arxiv:1910.10683",
"transformers",
"license:apache-2.0",
"text2text-generation"
]
| text2text-generation | [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin",
"special_tokens_map.json",
"tf_model.h5",
"tokenizer_config.json"
]
| google | 955 | transformers | ---
language: multilingual
datasets:
- mc4
license: apache-2.0
---
# ByT5 - Base
ByT5 is a tokenizer-free version of [Google's T5](https://ai.googleblog.com/2020/02/exploring-transfer-learning-with-t5.html) and generally follows the architecture of [MT5](https://huggingface.co/google/mt5-base).
ByT5 was only pre-trained on [mC4](https://www.tensorflow.org/datasets/catalog/c4#c4multilingual) excluding any supervised training with an average span-mask of 20 UTF-8 characters. Therefore, this model has to be fine-tuned before it is useable on a downstream task.
ByT5 works especially well on noisy text data,*e.g.*, `google/byt5-base` significantly outperforms [mt5-base](https://huggingface.co/google/mt5-base) on [TweetQA](https://arxiv.org/abs/1907.06292).
Paper: [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/pdf/1910.10683.pdf)
Authors: *Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel*
## Example Inference
ByT5 works on raw UTF-8 bytes and can be used without a tokenizer:
```python
from transformers import T5ForConditionalGeneration
import torch
model = T5ForConditionalGeneration.from_pretrained('google/byt5-base')
input_ids = torch.tensor([list("Life is like a box of chocolates.".encode("utf-8"))]) + 3 # add 3 for special tokens
labels = torch.tensor([list("La vie est comme une boîte de chocolat.".encode("utf-8"))]) + 3 # add 3 for special tokens
loss = model(input_ids, labels=labels).loss # forward pass
```
For batched inference & training it is however recommended using a tokenizer class for padding:
```python
from transformers import T5ForConditionalGeneration, AutoTokenizer
model = T5ForConditionalGeneration.from_pretrained('google/byt5-base')
tokenizer = AutoTokenizer.from_pretrained('google/byt5-base')
model_inputs = tokenizer(["Life is like a box of chocolates.", "Today is Monday."], padding="longest", return_tensors="pt")
labels = tokenizer(["La vie est comme une boîte de chocolat.", "Aujourd'hui c'est lundi."], padding="longest", return_tensors="pt").input_ids
loss = model(**model_inputs, labels=labels).loss # forward pass
```
## Abstract
Most widely-used pre-trained language models operate on sequences of tokens corresponding to word or subword units. Encoding text as a sequence of tokens requires a tokenizer, which is typically created as an independent artifact from the model. Token-free models that instead operate directly on raw text (bytes or characters) have many benefits: they can process text in any language out of the box, they are more robust to noise, and they minimize technical debt by removing complex and error-prone text preprocessing pipelines. Since byte or character sequences are longer than token sequences, past work on token-free models has often introduced new model architectures designed to amortize the cost of operating directly on raw text. In this paper, we show that a standard Transformer architecture can be used with minimal modifications to process byte sequences. We carefully characterize the trade-offs in terms of parameter count, training FLOPs, and inference speed, and show that byte-level models are competitive with their token-level counterparts. We also demonstrate that byte-level models are significantly more robust to noise and perform better on tasks that are sensitive to spelling and pronunciation. As part of our contribution, we release a new set of pre-trained byte-level Transformer models based on the T5 architecture, as well as all code and data used in our experiments.
 |
google/byt5-large | 2021-06-01T17:51:17.000Z | [
"pytorch",
"tf",
"t5",
"seq2seq",
"multilingual",
"dataset:mc4",
"arxiv:1907.06292",
"arxiv:1910.10683",
"transformers",
"license:apache-2.0",
"text2text-generation"
]
| text2text-generation | [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin",
"special_tokens_map.json",
"tf_model.h5",
"tokenizer_config.json"
]
| google | 291 | transformers | ---
language: multilingual
datasets:
- mc4
license: apache-2.0
---
# ByT5 - large
ByT5 is a tokenizer-free version of [Google's T5](https://ai.googleblog.com/2020/02/exploring-transfer-learning-with-t5.html) and generally follows the architecture of [MT5](https://huggingface.co/google/mt5-large).
ByT5 was only pre-trained on [mC4](https://www.tensorflow.org/datasets/catalog/c4#c4multilingual) excluding any supervised training with an average span-mask of 20 UTF-8 characters. Therefore, this model has to be fine-tuned before it is useable on a downstream task.
ByT5 works especially well on noisy text data,*e.g.*, `google/byt5-large` significantly outperforms [mt5-large](https://huggingface.co/google/mt5-large) on [TweetQA](https://arxiv.org/abs/1907.06292).
Paper: [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/pdf/1910.10683.pdf)
Authors: *Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel*
## Example Inference
ByT5 works on raw UTF-8 bytes and can be used without a tokenizer:
```python
from transformers import T5ForConditionalGeneration
import torch
model = T5ForConditionalGeneration.from_pretrained('google/byt5-large')
input_ids = torch.tensor([list("Life is like a box of chocolates.".encode("utf-8"))]) + 3 # add 3 for special tokens
labels = torch.tensor([list("La vie est comme une boîte de chocolat.".encode("utf-8"))]) + 3 # add 3 for special tokens
loss = model(input_ids, labels=labels).loss # forward pass
```
For batched inference & training it is however recommended using a tokenizer class for padding:
```python
from transformers import T5ForConditionalGeneration, AutoTokenizer
model = T5ForConditionalGeneration.from_pretrained('google/byt5-large')
tokenizer = AutoTokenizer.from_pretrained('google/byt5-large')
model_inputs = tokenizer(["Life is like a box of chocolates.", "Today is Monday."], padding="longest", return_tensors="pt")
labels = tokenizer(["La vie est comme une boîte de chocolat.", "Aujourd'hui c'est lundi."], padding="longest", return_tensors="pt").input_ids
loss = model(**model_inputs, labels=labels).loss # forward pass
```
## Abstract
Most widely-used pre-trained language models operate on sequences of tokens corresponding to word or subword units. Encoding text as a sequence of tokens requires a tokenizer, which is typically created as an independent artifact from the model. Token-free models that instead operate directly on raw text (bytes or characters) have many benefits: they can process text in any language out of the box, they are more robust to noise, and they minimize technical debt by removing complex and error-prone text preprocessing pipelines. Since byte or character sequences are longer than token sequences, past work on token-free models has often introduced new model architectures designed to amortize the cost of operating directly on raw text. In this paper, we show that a standard Transformer architecture can be used with minimal modifications to process byte sequences. We carefully characterize the trade-offs in terms of parameter count, training FLOPs, and inference speed, and show that byte-level models are competitive with their token-level counterparts. We also demonstrate that byte-level models are significantly more robust to noise and perform better on tasks that are sensitive to spelling and pronunciation. As part of our contribution, we release a new set of pre-trained byte-level Transformer models based on the T5 architecture, as well as all code and data used in our experiments.
 |
google/byt5-small | 2021-06-01T17:50:58.000Z | [
"pytorch",
"tf",
"t5",
"seq2seq",
"multilingual",
"dataset:mc4",
"arxiv:1907.06292",
"arxiv:1910.10683",
"transformers",
"license:apache-2.0",
"text2text-generation"
]
| text2text-generation | [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin",
"special_tokens_map.json",
"tf_model.h5",
"tokenizer_config.json"
]
| google | 2,308 | transformers | ---
language: multilingual
datasets:
- mc4
license: apache-2.0
---
# ByT5 - Small
ByT5 is a tokenizer-free version of [Google's T5](https://ai.googleblog.com/2020/02/exploring-transfer-learning-with-t5.html) and generally follows the architecture of [MT5](https://huggingface.co/google/mt5-small).
ByT5 was only pre-trained on [mC4](https://www.tensorflow.org/datasets/catalog/c4#c4multilingual) excluding any supervised training with an average span-mask of 20 UTF-8 characters. Therefore, this model has to be fine-tuned before it is useable on a downstream task.
ByT5 works especially well on noisy text data,*e.g.*, `google/byt5-small` significantly outperforms [mt5-small](https://huggingface.co/google/mt5-small) on [TweetQA](https://arxiv.org/abs/1907.06292).
Paper: [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/pdf/1910.10683.pdf)
Authors: *Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel*
## Example Inference
ByT5 works on raw UTF-8 bytes and can be used without a tokenizer:
```python
from transformers import T5ForConditionalGeneration
import torch
model = T5ForConditionalGeneration.from_pretrained('google/byt5-small')
input_ids = torch.tensor([list("Life is like a box of chocolates.".encode("utf-8"))]) + 3 # add 3 for special tokens
labels = torch.tensor([list("La vie est comme une boîte de chocolat.".encode("utf-8"))]) + 3 # add 3 for special tokens
loss = model(input_ids, labels=labels).loss # forward pass
```
For batched inference & training it is however recommended using a tokenizer class for padding:
```python
from transformers import T5ForConditionalGeneration, AutoTokenizer
model = T5ForConditionalGeneration.from_pretrained('google/byt5-small')
tokenizer = AutoTokenizer.from_pretrained('google/byt5-small')
model_inputs = tokenizer(["Life is like a box of chocolates.", "Today is Monday."], padding="longest", return_tensors="pt")
labels = tokenizer(["La vie est comme une boîte de chocolat.", "Aujourd'hui c'est lundi."], padding="longest", return_tensors="pt").input_ids
loss = model(**model_inputs, labels=labels).loss # forward pass
```
## Abstract
Most widely-used pre-trained language models operate on sequences of tokens corresponding to word or subword units. Encoding text as a sequence of tokens requires a tokenizer, which is typically created as an independent artifact from the model. Token-free models that instead operate directly on raw text (bytes or characters) have many benefits: they can process text in any language out of the box, they are more robust to noise, and they minimize technical debt by removing complex and error-prone text preprocessing pipelines. Since byte or character sequences are longer than token sequences, past work on token-free models has often introduced new model architectures designed to amortize the cost of operating directly on raw text. In this paper, we show that a standard Transformer architecture can be used with minimal modifications to process byte sequences. We carefully characterize the trade-offs in terms of parameter count, training FLOPs, and inference speed, and show that byte-level models are competitive with their token-level counterparts. We also demonstrate that byte-level models are significantly more robust to noise and perform better on tasks that are sensitive to spelling and pronunciation. As part of our contribution, we release a new set of pre-trained byte-level Transformer models based on the T5 architecture, as well as all code and data used in our experiments.
|
google/byt5-xl | 2021-06-14T19:45:15.000Z | [
"pytorch",
"tf",
"t5",
"seq2seq",
"multilingual",
"dataset:mc4",
"arxiv:1907.06292",
"arxiv:1910.10683",
"transformers",
"license:apache-2.0",
"text2text-generation"
]
| text2text-generation | [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin",
"special_tokens_map.json",
"tf_model.h5",
"tokenizer_config.json"
]
| google | 306 | transformers | ---
language: multilingual
datasets:
- mc4
license: apache-2.0
---
# ByT5 - xl
ByT5 is a tokenizer-free version of [Google's T5](https://ai.googleblog.com/2020/02/exploring-transfer-learning-with-t5.html) and generally follows the architecture of [MT5](https://huggingface.co/google/mt5-xl).
ByT5 was only pre-trained on [mC4](https://www.tensorflow.org/datasets/catalog/c4#c4multilingual) excluding any supervised training with an average span-mask of 20 UTF-8 characters. Therefore, this model has to be fine-tuned before it is useable on a downstream task.
ByT5 works especially well on noisy text data,*e.g.*, `google/byt5-xl` significantly outperforms [mt5-xl](https://huggingface.co/google/mt5-xl) on [TweetQA](https://arxiv.org/abs/1907.06292).
Paper: [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/pdf/1910.10683.pdf)
Authors: *Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel*
## Example Inference
ByT5 works on raw UTF-8 bytes and can be used without a tokenizer:
```python
from transformers import T5ForConditionalGeneration
import torch
model = T5ForConditionalGeneration.from_pretrained('google/byt5-xl')
input_ids = torch.tensor([list("Life is like a box of chocolates.".encode("utf-8"))]) + 3 # add 3 for special tokens
labels = torch.tensor([list("La vie est comme une boîte de chocolat.".encode("utf-8"))]) + 3 # add 3 for special tokens
loss = model(input_ids, labels=labels).loss # forward pass
```
For batched inference & training it is however recommended using a tokenizer class for padding:
```python
from transformers import T5ForConditionalGeneration, AutoTokenizer
model = T5ForConditionalGeneration.from_pretrained('google/byt5-xl')
tokenizer = AutoTokenizer.from_pretrained('google/byt5-xl')
model_inputs = tokenizer(["Life is like a box of chocolates.", "Today is Monday."], padding="longest", return_tensors="pt")
labels = tokenizer(["La vie est comme une boîte de chocolat.", "Aujourd'hui c'est lundi."], padding="longest", return_tensors="pt").input_ids
loss = model(**model_inputs, labels=labels).loss # forward pass
```
## Abstract
Most widely-used pre-trained language models operate on sequences of tokens corresponding to word or subword units. Encoding text as a sequence of tokens requires a tokenizer, which is typically created as an independent artifact from the model. Token-free models that instead operate directly on raw text (bytes or characters) have many benefits: they can process text in any language out of the box, they are more robust to noise, and they minimize technical debt by removing complex and error-prone text preprocessing pipelines. Since byte or character sequences are longer than token sequences, past work on token-free models has often introduced new model architectures designed to amortize the cost of operating directly on raw text. In this paper, we show that a standard Transformer architecture can be used with minimal modifications to process byte sequences. We carefully characterize the trade-offs in terms of parameter count, training FLOPs, and inference speed, and show that byte-level models are competitive with their token-level counterparts. We also demonstrate that byte-level models are significantly more robust to noise and perform better on tasks that are sensitive to spelling and pronunciation. As part of our contribution, we release a new set of pre-trained byte-level Transformer models based on the T5 architecture, as well as all code and data used in our experiments.
 |
google/byt5-xxl | 2021-06-15T12:11:58.000Z | [
"pytorch",
"tf",
"t5",
"seq2seq",
"multilingual",
"dataset:mc4",
"arxiv:1907.06292",
"arxiv:1910.10683",
"transformers",
"license:apache-2.0",
"text2text-generation"
]
| text2text-generation | [
".gitattributes",
"README.md",
"config.json",
"pytorch_model.bin",
"special_tokens_map.json",
"tf_model.h5",
"tokenizer_config.json"
]
| google | 56 | transformers | ---
language: multilingual
datasets:
- mc4
license: apache-2.0
---
# ByT5 - xxl
ByT5 is a tokenizer-free version of [Google's T5](https://ai.googleblog.com/2020/02/exploring-transfer-learning-with-t5.html) and generally follows the architecture of [MT5](https://huggingface.co/google/mt5-xxl).
ByT5 was only pre-trained on [mC4](https://www.tensorflow.org/datasets/catalog/c4#c4multilingual) excluding any supervised training with an average span-mask of 20 UTF-8 characters. Therefore, this model has to be fine-tuned before it is useable on a downstream task.
ByT5 works especially well on noisy text data,*e.g.*, `google/byt5-xxl` significantly outperforms [mt5-xxl](https://huggingface.co/google/mt5-xxl) on [TweetQA](https://arxiv.org/abs/1907.06292).
Paper: [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/pdf/1910.10683.pdf)
Authors: *Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel*
## Example Inference
ByT5 works on raw UTF-8 bytes and can be used without a tokenizer:
```python
from transformers import T5ForConditionalGeneration
import torch
model = T5ForConditionalGeneration.from_pretrained('google/byt5-xxl')
input_ids = torch.tensor([list("Life is like a box of chocolates.".encode("utf-8"))]) + 3 # add 3 for special tokens
labels = torch.tensor([list("La vie est comme une boîte de chocolat.".encode("utf-8"))]) + 3 # add 3 for special tokens
loss = model(input_ids, labels=labels).loss # forward pass
```
For batched inference & training it is however recommended using a tokenizer class for padding:
```python
from transformers import T5ForConditionalGeneration, AutoTokenizer
model = T5ForConditionalGeneration.from_pretrained('google/byt5-xxl')
tokenizer = AutoTokenizer.from_pretrained('google/byt5-xxl')
model_inputs = tokenizer(["Life is like a box of chocolates.", "Today is Monday."], padding="longest", return_tensors="pt")
labels = tokenizer(["La vie est comme une boîte de chocolat.", "Aujourd'hui c'est lundi."], padding="longest", return_tensors="pt").input_ids
loss = model(**model_inputs, labels=labels).loss # forward pass
```
## Abstract
Most widely-used pre-trained language models operate on sequences of tokens corresponding to word or subword units. Encoding text as a sequence of tokens requires a tokenizer, which is typically created as an independent artifact from the model. Token-free models that instead operate directly on raw text (bytes or characters) have many benefits: they can process text in any language out of the box, they are more robust to noise, and they minimize technical debt by removing complex and error-prone text preprocessing pipelines. Since byte or character sequences are longer than token sequences, past work on token-free models has often introduced new model architectures designed to amortize the cost of operating directly on raw text. In this paper, we show that a standard Transformer architecture can be used with minimal modifications to process byte sequences. We carefully characterize the trade-offs in terms of parameter count, training FLOPs, and inference speed, and show that byte-level models are competitive with their token-level counterparts. We also demonstrate that byte-level models are significantly more robust to noise and perform better on tasks that are sensitive to spelling and pronunciation. As part of our contribution, we release a new set of pre-trained byte-level Transformer models based on the T5 architecture, as well as all code and data used in our experiments.
 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.