MicroGPT::ModelSpec
Parse a model spec like "d64n2", "d4x16n2ff512", or "d1+2+4+8+16+32n2" d64 = one head, dim 64 d4x16 = 4 uniform heads of dim 16 (d_model=64) d1+2+4 = heterogeneous heads (d_model=7)
Constructors
Instance methods
clone
Sourcecopy_with(d_model _d_model = @d_model, n_layers _n_layers = @n_layers, d_ff _d_ff = @d_ff, head_dims _head_dims = @head_dims)
Sourced_ff
d_model
head_dims
label
Sourcen_layers