LLM Zoo πŸ¦“

Transformer: 24 layersOutput Projection2880 β†’ 196KRMSNorm2880Mixture of Experts4 / 32 experts2880 β†’ 2088 β†’ 2880RMSNorm2880Grouped Query Attention64 heads, 8 KV (8:1)2880 β†’ 2880RMSNorm2880ConcatenateEmbedding196K β†’ 2880Vision EncoderViT, 14px patchesimage β†’ 2880Text (input)Image (input)