llama

type Layer struct {
	AttentionNorm *nn.RMSNorm `gguf:"attn_norm"`
	SelfAttention *SelfAttention
	MLPNorm       *nn.RMSNorm `gguf:"ffn_norm"`
	MLP           *MLP
}

type MLP struct {
	Up   *nn.Linear `gguf:"ffn_up"`
	Down *nn.Linear `gguf:"ffn_down"`
	Gate *nn.Linear `gguf:"ffn_gate"`
}

type Model struct {
	model.Base
	model.TextProcessor

	TokenEmbedding *nn.Embedding `gguf:"token_embd"`
	Layers         []Layer       `gguf:"blk"`
	OutputNorm     *nn.RMSNorm   `gguf:"output_norm"`
	Output         *nn.Linear    `gguf:"output,alt:token_embd"`

	Options
}

type Options struct {
	// contains filtered or unexported fields
}

type SelfAttention struct {
	Query       *nn.Linear `gguf:"attn_q"`
	Key         *nn.Linear `gguf:"attn_k"`
	Value       *nn.Linear `gguf:"attn_v"`
	Output      *nn.Linear `gguf:"attn_output"`
	RopeFactors ml.Tensor  `gguf:"rope_freqs.weight"`
}