LLM Zoo
π¦
Info
Γ24
layers
Transformer: 24 layers
Output Projection
2880 β 196K
RMSNorm
2880
Mixture of Experts
4 / 32 experts
2880 β 2088 β 2880
RMSNorm
2880
Grouped Query Attention
64 heads, 8 KV (8:1)
2880 β 2880
RMSNorm
2880
Concatenate
Embedding
196K β 2880
Vision Encoder
ViT, 14px patches
image β 2880
Text (input)
Image (input)