← MIC coverage of Transformers model types

mistralai/Mistral-7B-Instruct-v0.2

huggingface-mic/Mistral-7B-Instruct-v0.2-mic@90938461b4cb ← mistralai/Mistral-7B-Instruct-v0.2@63a8b0818953 · MistralForCausalLM · random weights · built before the shrink rule (no recorded recipe)

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

Original7.2Bparameters
Fixture8.4Mparameters
Layer stacks1compared
Findings1to review

⚠ To review

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 · 32
Fixture · 2
KindOriginalFixture
DecoderLayer322✓ 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.layers7.0B96.4%246.3K2.9%
model.embed_tokens131.1M1.8%4.1M48.5%
lm_head131.1M1.8%4.1M48.5%
model.norm4.1K0.0%1280.0%

Preserved properties

ConfigPropertyOriginalFixture
modelattention styleGQA (32 heads, 8 KV)GQA (4 heads, 2 KV)✓ kept
modelRoPE typedefaultdefault✓ kept
modelmodel_typemistralmistral✓ kept
modelvocab_size3200032000✓ kept
modeltie_word_embeddingsFalseFalse✓ kept
modelhidden_actsilusilu✓ kept
modelmax_position_embeddings32768512⚠ changed

Module classes

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

ClassOriginalFixture
Linear22515
RMSNorm655
Attention322
DecoderLayer322
MLP322
SiLUActivation322
Embedding11
ForCausalLM11
Model11
RotaryEmbedding11
ModuleList11

Config changes

7 fields changed, 21 unchanged.

FieldOriginalFixture
head_dim12816
hidden_size4096128
intermediate_size14336256
max_position_embeddings32768512
num_attention_heads324
num_hidden_layers322
num_key_value_heads82