Module architectures 

Model architecture implementations

Re-exports

pub use bert::BertModelWrapper;

pub use clip::ClipModelWrapper;

pub use llama::LlamaModelWrapper;

pub use qwen2::Qwen2ModelWrapper;

pub use qwen3::Qwen3ModelWrapper;

Modules

bert

BERT architecture using Candle’s built-in implementation BERT is an encoder model used for embeddings and classification tasks

clip

CLIP model wrapper — supports OpenAI CLIP, Chinese-CLIP, and SigLIP.

llama

Custom Llama architecture with public fields and per-request KV cache.

qwen2

Qwen2 architecture using Candle’s built-in implementation

qwen3

Qwen3 architecture implementation