Module nomic_moe 

Nomic V2 MoE embedding provider via fastembed-rs candle backend.

Uses the candle-nn pure Rust inference engine (no ONNX Runtime). Models are downloaded from HuggingFace on first use.

nomic-ai/nomic-embed-text-v2-moe — 475M total / 305M active params, 8 experts with top-2 routing, hidden_size=768.

use llm_kernel::embedding::NomicMoeProvider;
use llm_kernel::embedding::EmbeddingProvider;

let provider = NomicMoeProvider::new()?;
let result = provider.embed("hello world")?;

Structs

NomicMoeProvider

Nomic V2 MoE embedding provider backed by candle-nn.

Constants

NOMIC_EMBED_TEXT_V2_MOE

Default HuggingFace repo for nomic-embed-text-v2-moe.