Enum EmbeddingModel 

pub enum EmbeddingModel {
    BgeLarge,
    MiniLM,
    BgeSmall,
    E5Small,
    ModernBertEmbedBase,
}

Supported embedding models.

Variants

BgeLarge

BAAI/bge-large-en-v1.5 - Highest quality, 1024 dimensions (default)

• Dimensions: 1024
• Max tokens: 512
• Speed: Slower than small models, but highest quality

MiniLM

all-MiniLM-L6-v2 - Fast and efficient, good for general use

• Dimensions: 384
• Max tokens: 256
• Speed: Fastest

BgeSmall

BAAI/bge-small-en-v1.5 - Balanced quality and speed

• Dimensions: 384
• Max tokens: 512
• Speed: Medium

E5Small

intfloat/e5-small-v2 - Higher quality embeddings

• Dimensions: 384
• Max tokens: 512
• Speed: Medium

ModernBertEmbedBase

nomic-ai/modernbert-embed-base — modern transformer with 8192-token context

• Dimensions: 768 (native Matryoshka: 768/512/256/128/64)
• Max tokens: 8192
• Speed: 25% faster than BGE-Large on same hardware
• Flash Attention support for long sequences
• Env var: DAKERA_MODEL=modernbert-embed-base

Implementations

impl EmbeddingModel

pub fn model_id(&self) -> &'static str

Get the HuggingFace model ID.

pub fn dimension(&self) -> usize

Get the embedding dimension for this model.

pub fn max_seq_length(&self) -> usize

Get the maximum sequence length (in tokens).

pub fn mrl_dimensions(&self) -> Option<&'static [usize]>

Get the Matryoshka-supported dimensions for this model (smallest to largest).

Returns None for models that do not support MRL truncation.

pub fn safetensors_filename(&self) -> &'static str

Get the safetensors model filename (for Candle backend).

pub fn config_filename(&self) -> &'static str

Get the config filename (for Candle/GGUF backends).

pub fn model2vec_repo_id(&self) -> &'static str

Get the HuggingFace repo hosting the Model2Vec distilled vocabulary matrix.

pub fn gguf_repo_id(&self) -> &'static str

Get the HuggingFace repo hosting the GGUF quantised models for this embedding.

pub fn query_prefix(&self) -> Option<&'static str>

Get the query prefix for models that require it. Some models like E5 require a prefix for queries vs documents.

pub fn document_prefix(&self) -> Option<&'static str>

Get the document/passage prefix for models that require it.

pub fn use_mean_pooling(&self) -> bool

Whether this model uses mean pooling (vs CLS token).

pub fn normalize_embeddings(&self) -> bool

Whether embeddings should be normalized.

pub fn tokens_per_second_cpu(&self) -> usize

Get approximate tokens per second on CPU (for estimation).

pub fn onnx_repo_id(&self) -> &'static str

Get the HuggingFace repository ID hosting the ONNX INT8 model for this embedding model.

These are Xenova-hosted Optimum ONNX exports — quantized INT8, pre-built, no conversion needed. BgeLarge: ~130 MB, MiniLM: 23 MB, BGE-small: 35 MB, E5-small: 35 MB.

pub fn onnx_filename(&self) -> &'static str

Get the ONNX model filename for CPU inference (INT8 quantized).

pub fn onnx_filename_gpu(&self) -> &'static str

Get the ONNX model filename for GPU (CUDA EP) inference.

Returns the FP32 model (onnx/model.onnx) instead of INT8. The INT8 quantized model has 336 Memcpy CPU↔GPU round-trips caused by ORT falling back to CPU EP for every unsupported INT8 op — making CUDA 24× slower than pure CPU inference. The FP32 model contains no unsupported ops and runs entirely on-device.

pub fn all() -> &'static [EmbeddingModel]

List all available models.

pub fn parse(s: &str) -> Option<Self>

Parse model from string (case-insensitive).

pub fn from_env() -> Self

Get the active model from DAKERA_MODEL env var, defaulting to BgeLarge.

Trait Implementations

impl Clone for EmbeddingModel

fn clone(&self) -> EmbeddingModel

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for EmbeddingModel

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for EmbeddingModel

fn default() -> EmbeddingModel

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for EmbeddingModel

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Display for EmbeddingModel

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Hash for EmbeddingModel

fn hash<__H: Hasher>(&self, state: &mut __H)

Feeds this value into the given Hasher.

fn hash_slice<H>(data: &[Self], state: &mut H)

where H: Hasher, Self: Sized,

Feeds a slice of this type into the given Hasher.

impl PartialEq for EmbeddingModel

fn eq(&self, other: &EmbeddingModel) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Serialize for EmbeddingModel

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl Copy for EmbeddingModel

impl Eq for EmbeddingModel

impl StructuralPartialEq for EmbeddingModel