← MIC coverage of Transformers model types

Salesforce/blip-image-captioning-base

huggingface-mic/blip-image-captioning-base-mic@065eddd4b404 ← Salesforce/blip-image-captioning-base@82a37760796d · BlipForConditionalGeneration · random weights

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

Original224.0Mparameters
Fixture2.3Mparameters
Layer stacks2compared
Findings0to review

✓ Same architecture, smaller

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

Model tester (advisory)

The model's own small test config in its ModelTester (19 literal values, 19 set by the model's own tester, the rest generic base-tester defaults). Testers are usually representative but not always, so these notes are for review, not rules.

The fixture covers every per-layer kind the tester sets, and shrinks every size it does.

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.

vision_model.encoder.layers

Original · 12
Fixture · 2
KindOriginalFixture
EncoderLayer122✓ kept

text_decoder.bert.encoder.layer

Original · 12
Fixture · 2
KindOriginalFixture
TextLayer122✓ 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
text_decoder.bert.encoder.layer113.4M50.6%100.5K4.4%
vision_model.encoder.layers85.1M38.0%66.9K2.9%
text_decoder.bert23.8M10.6%2.0M87.3%
vision_model.embeddings1.0M0.5%86.2K3.8%
text_decoder.cls622.7K0.3%34.8K1.5%
vision_model.post_layernorm1.5K0.0%1280.0%

Preserved properties

ConfigPropertyOriginalFixture
modelmodel_typeblipblip✓ kept
modeltie_word_embeddingsTrueTrue✓ kept
text_configmodel_typeblip_text_modelblip_text_model✓ kept
text_configvocab_size3052430524✓ kept
text_configtie_word_embeddingsTrueTrue✓ kept
text_confighidden_actgelugelu✓ kept
text_configmax_position_embeddings512512✓ kept
vision_configmodel_typeblip_vision_modelblip_vision_model✓ kept
vision_configtie_word_embeddingsTrueTrue✓ kept
vision_confighidden_actgelugelu✓ kept
vision_configimage_size384384✓ kept
vision_configpatch_size1616✓ kept
vision_confignum_channels33✓ kept

Module classes

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

ClassOriginalFixture
Linear17030
Dropout7313
LayerNorm6313
GELUActivation255
TextAttention244
TextSelfAttention244
TextSelfOutput244
Attention122
EncoderLayer122
MLP122
TextIntermediate122
TextLayer122
TextOutput122
Embedding22
ModuleList22
Encoder11
ForConditionalGeneration11
TextEmbeddings11
TextEncoder11
TextLMHeadModel11
TextLMPredictionHead11
TextModel11
TextOnlyMLMHead11
TextPredictionHeadTransform11
VisionEmbeddings11
VisionModel11
Conv2d11

Config changes

12 fields changed, 88 unchanged.

FieldOriginalFixture
projection_dim51232
text_config.encoder_hidden_size76864
text_config.hidden_size76864
text_config.intermediate_size3072128
text_config.num_attention_heads124
text_config.num_hidden_layers122
text_config.projection_dim76832
vision_config.hidden_size76864
vision_config.intermediate_size3072128
vision_config.num_attention_heads124
vision_config.num_hidden_layers122
vision_config.projection_dim51232