← MIC coverage of Transformers model types

facebook/hubert-large-ls960-ft

huggingface-mic/hubert-large-ls960-ft-mic@5f4ea3f40f37 ← facebook/hubert-large-ls960-ft@ece5fabbf034 · HubertForCTC · overfit weights · built before the shrink rule (no recorded recipe)

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

Original315.5Mparameters
Fixture45.0Kparameters
Layer stacks2compared
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.

hubert.feature_extractor.conv_layers

Original · 7
Fixture · 7
KindOriginalFixture
LayerNormConvLayer77✓ kept

hubert.encoder.layers

Original · 24
Fixture · 2
KindOriginalFixture
Linear242✓ 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
hubert.encoder.layers302.3M95.8%17.1K38.0%
hubert.encoder8.4M2.7%8.3K18.5%
hubert.feature_extractor.conv_layers4.2M1.3%17.4K38.6%
hubert.feature_projection526.3K0.2%1.1K2.5%
lm_head32.8K0.0%1.1K2.3%
hubert1.0K0.0%320.1%

Preserved properties

ConfigPropertyOriginalFixture
modelmodel_typehuberthubert✓ kept
modelvocab_size3232✓ kept
modelhidden_actgelugelu✓ kept

Module classes

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

ClassOriginalFixture
Linear14614
Dropout759
LayerNorm5713
GELUActivation3210
Attention242
EncoderLayerStableLayerNorm242
FeedForward242
Conv1d77
LayerNormConvLayer77
ModuleList22
EncoderStableLayerNorm11
FeatureEncoder11
FeatureProjection11
ForCTC11
Model11
PositionalConvEmbedding11
SamePadLayer11
ModuleDict11
ParametrizationList11
ParametrizedConv1d11
_WeightNorm11

Config changes

7 fields changed, 46 unchanged.

FieldOriginalFixture
conv_dim7 entries: 7× 5127 entries: 7× 32
hidden_size102432
intermediate_size409664
num_attention_heads164
num_conv_pos_embedding_groups164
num_conv_pos_embeddings12832
num_hidden_layers242