pub(super) fn embed_sparse(
session: &mut Session,
tokenizer: &Tokenizer,
texts: &[String],
cost_model: &CostModel,
model_variant: ModelVariant,
) -> Result<(Vec<SparseEmbedding>, EmbedStats)>Expand description
Produces sparse embeddings via the BGE-M3 sparse-linear projection.
Tokenizes once, then uses the cost model to bin-pack into chunks. Results are scattered back to the original input order.