← MIC coverage of Transformers model types

Qwen/Qwen2-0.5B

huggingface-mic/qwen2-mic@f9c74f62c625 ← Qwen/Qwen2-0.5B@91d2aff3f957 · Qwen2ForCausalLM · random weights · built before the shrink rule (no recorded recipe)

CI: passing now; 2 failed checks in 30 days · MIC failures dashboard

Original494.0Mparameters
Fixture9.8Mparameters
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.

model.layers

Original · 24
Fixture · 2
KindOriginalFixture
DecoderLayer242✓ 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.layers357.9M72.4%74.2K0.8%
model.embed_tokens136.1M27.6%9.7M99.2%
model.norm8960.0%640.0%

Preserved properties

ConfigPropertyOriginalFixture
modelattention styleGQA (14 heads, 2 KV)GQA (4 heads, 2 KV)✓ kept
modelRoPE typedefaultdefault✓ kept
modelmodel_typeqwen2qwen2✓ kept
modelvocab_size151936151936✓ kept
modeltie_word_embeddingsTrueTrue✓ kept
modelhidden_actsilusilu✓ kept
modelmax_position_embeddings131072131072✓ kept

Module classes

Instances of each class in the built model (prefix Qwen2 omitted).

ClassOriginalFixture
Linear16915
RMSNorm495
Attention242
DecoderLayer242
MLP242
SiLUActivation242
Embedding11
ModuleList11
ForCausalLM11
Model11
RotaryEmbedding11

Config changes

5 fields changed, 25 unchanged.

FieldOriginalFixture
hidden_size89664
intermediate_size4864128
layer_types24 entries: 24× full_attention["full_attention", "full_attention"]
num_attention_heads144
num_hidden_layers242