← MIC coverage of Transformers model types

microsoft/swinv2-tiny-patch4-window16-256

huggingface-mic/swinv2-tiny-patch4-window16-256-mic@8c997579b76f ← microsoft/swinv2-tiny-patch4-window16-256@f4d3075206f2 · Swinv2ForImageClassification · random weights

CI: no failure on main in 30 days · MIC failures dashboard

Original28.3Mparameters
Fixture5.4Mparameters
Layer stacks1compared
Findings0to review

✓ Same architecture, smaller

Every module class, layer kind and checked property of the original is in the fixture.

Layer stacks

One square per layer, colored by kind: the layer's module classes and type attributes. Hover a square for its index and kind.

swinv2.encoder.layers

Original · 4
Fixture · 4
KindOriginalFixture
PatchMerging33✓ kept
no PatchMerging11✓ kept

Where the parameters are

Share of each model's parameters per component. Vocabulary and patch sizes are kept, so embeddings usually take a larger share of the fixture.

Original
Fixture
ComponentOriginalShareFixtureShare
swinv2.encoder.layers27.6M97.3%4.9M90.5%
classifier769.0K2.7%513.0K9.5%
swinv2.embeddings4.9K0.0%3.3K0.1%
swinv2.layernorm1.5K0.0%1.0K0.0%

Preserved properties

ConfigPropertyOriginalFixture
modelmodel_typeswinv2swinv2✓ kept
modelhidden_actgelugelu✓ kept
modelimage_size256256✓ kept
modelpatch_size44✓ kept
modelnum_channels33✓ kept

Module classes

Instances of each class in the built model (prefix Swinv2 omitted).

ClassOriginalFixture
Linear10036
Dropout3713
LayerNorm2913
GELUActivation124
ReLU124
Sequential124
Attention124
Intermediate124
Layer124
Output124
SelfAttention124
SelfOutput124
DropPath113
ModuleList55
Stage44
PatchMerging33
AdaptiveAvgPool1d11
Conv2d11
Identity11
Embeddings11
Encoder11
ForImageClassification11
Model11
PatchEmbeddings11

Config changes

5 fields changed, 28 unchanged.

FieldOriginalFixture
depths[2, 2, 6, 2][1, 1, 1, 1]
embed_dim9664
hidden_size76864
num_heads[3, 6, 12, 24][2, 4, 8, 16]
num_layers42