hustvl/yolos-small

Original30.7Mparameters

Fixture379.7Kparameters

Layer stacks1compared

Findings0to review

✓ Same architecture, smaller

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

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

Original · 12

Fixture · 2

Kind	Original	Fixture
Layer	12	2	✓ kept

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

Component	Original	Share	Fixture	Share
vit.encoder.layer	21.3M	69.4%	66.9K	17.6%
vit.encoder	7.7M	25.2%	117.1K	30.8%
vit.embeddings	1.0M	3.4%	172.7K	45.5%
class_labels_classifier.layers	331.1K	1.1%	14.3K	3.8%
bbox_predictor.layers	297.2K	1.0%	8.6K	2.3%
vit.layernorm	768	0.0%	128	0.0%

Config	Property	Original	Fixture
model	model_type	yolos	yolos	✓ kept
model	hidden_act	gelu	gelu	✓ kept
model	image_size	[512, 864]	[512, 864]	✓ kept
model	patch_size	16	16	✓ kept
model	num_channels	3	3	✓ kept

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

Class	Original	Fixture
Linear	78	18
Dropout	25	5
LayerNorm	25	5
GELUActivation	12	2
Attention	12	2
Intermediate	12	2
Layer	12	2
Output	12	2
SelfAttention	12	2
SelfOutput	12	2
ModuleList	3	3
MLPPredictionHead	2	2
Conv2d	1	1
InterpolateInitialPositionEmbeddings	1	1
InterpolateMidPositionEmbeddings	1	1
Embeddings	1	1
Encoder	1	1
ForObjectDetection	1	1
Model	1	1
PatchEmbeddings	1	1

4 fields changed, 27 unchanged.