Crate llama_models 

llama-models

Foundational model blocks for llama.rs:

• RMSNorm
• RoPE
• Attention (scaled dot-product, single-step decode form)
• MLP (SwiGLU)
• Safetensors weight loading

Structs

LlamaBlock

Minimal Llama block composition.

ModelWeights

Named weight storage loaded from safetensors.

QwenBlock

Minimal Qwen block composition (same block primitives at this stage).

Tensor

Lightweight tensor holder for loaded model weights.

Enums

ModelError

Errors for model operations and weight loading.

Functions

apply_rope

Apply rotary positional embeddings in-place to query and key vectors.

attention_decode

Single-step decode attention:

mlp_swiglu

SwiGLU MLP:

rms_norm

Root mean square normalization.