Struct ClassicBertWeights 

pub struct ClassicBertWeights<T> {
    pub word_embeddings: T,
    pub position_embeddings: T,
    pub token_type_embeddings: T,
    pub emb_ln_weight: T,
    pub emb_ln_bias: T,
    pub layers: Vec<ClassicBertLayerWeights<T>>,
    pub num_heads: usize,
    pub head_dim: usize,
    pub hidden_dim: usize,
    pub intermediate_dim: usize,
    pub layer_norm_eps: f32,
}

Full ClassicBert model weights, generic over tensor type.

Includes embedding tables, per-layer encoder weights, and model geometry. The tensor type T becomes Driver::Tensor when loaded onto a specific backend.

Fields

word_embeddings: T

Word embedding table [vocab_size, hidden].

position_embeddings: T

Learned position embedding table [max_position, hidden].

token_type_embeddings: T

Token type embedding table [2, hidden].

emb_ln_weight: T

Post-embedding LayerNorm weight [hidden].

emb_ln_bias: T

Post-embedding LayerNorm bias [hidden].

layers: Vec<ClassicBertLayerWeights<T>>

Per-layer encoder weights.

num_heads: usize

Number of attention heads (e.g., 12 for BGE-small).

head_dim: usize

Dimension per attention head (hidden / num_heads).

hidden_dim: usize

Hidden dimension (e.g., 384 for BGE-small).

intermediate_dim: usize

FFN intermediate dimension (e.g., 1536 for BGE-small).

layer_norm_eps: f32

Layer normalization epsilon (typically 1e-12).