← MIC coverage of Transformers model types

google/pegasus-xsum

huggingface-mic/pegasus-xsum-mic@3703172718d0 ← google/pegasus-xsum@8d8ffc158a3b · PegasusForConditionalGeneration · random weights

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

Original569.7Mparameters
Fixture6.4Mparameters
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.

model.encoder.layers

Original · 16
Fixture · 2
KindOriginalFixture
EncoderLayer162✓ kept

model.decoder.layers

Original · 16
Fixture · 2
KindOriginalFixture
DecoderLayer162✓ 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
model.decoder.layers268.7M47.2%100.5K1.6%
model.encoder.layers201.5M35.4%66.9K1.0%
model.shared98.4M17.3%6.2M96.3%
model.decoder526.3K0.1%32.9K0.5%
model.encoder526.3K0.1%32.9K0.5%

Preserved properties

ConfigPropertyOriginalFixture
modelmodel_typepegasuspegasus✓ kept
modelvocab_size9610396103✓ kept
modeltie_word_embeddingsTrueTrue✓ kept
modelmax_position_embeddings512512✓ kept

Module classes

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

ClassOriginalFixture
Linear25733
LayerNorm8212
Attention486
ReLU324
DecoderLayer162
EncoderLayer162
Embedding33
ModuleList22
SinusoidalPositionalEmbedding22
Decoder11
Encoder11
ForConditionalGeneration11
Model11

Config changes

7 fields changed, 38 unchanged.

FieldOriginalFixture
d_model102464
decoder_attention_heads164
decoder_ffn_dim4096128
decoder_layers162
encoder_attention_heads164
encoder_ffn_dim4096128
encoder_layers162