candding 0.1.0

Pure candle embeddings for Rust with no ONNX: dense, sparse, multi-vector and rerank models today, image models on the roadmap
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

//! The flash XLM-RoBERTa encoder of jina-colbert-v2 and jina-reranker-v2: fused QKV, post-norm blocks, and
//! rotary or learned positions per the entry's `PositionScheme`. No `model_type`, so only the registry reaches it.

mod attention;
mod config;
mod embeddings;
mod layer;
mod model;
mod position;

pub use config::XlmRobertaFlashConfig;
pub use model::XlmRobertaFlashModel;
pub use position::PositionScheme;

/// `Family::XlmRobertaFlash`'s serde name; no checkpoint publishes a `model_type`, so only the registry passes it.
pub const ARCHITECTURE: &str = "xlm_roberta_flash";