Enum PoolingType 

pub enum PoolingType {
    None,
    Mean,
    Cls,
    Last,
    Rank,
}

enum llama_pooling_type, by its wire values.

Variants

None

No pooling: every token’s hidden state is returned.

Mean

Arithmetic mean over the sequence.

Cls

The first token’s row. What BERT/BGE checkpoints use, because their [CLS] position is the one the model was trained to put the sentence representation in.

Last

The last token’s row. Decoder-style embedding models.

Rank

Not pooling — a reranker classification head. See module docs.

Implementations

impl PoolingType

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

The name upstream prints, so a refusal names the same thing the user’s llama-embedding --pooling flag does.

pub fn from_gguf( file: &impl TensorSource, arch: &str, ) -> Result<Option<Self>, PoolingError>

Reads {arch}.pooling_type. Ok(None) means the key is absent or explicitly unspecified — the caller picks a default and says so, rather than this function inventing one.

Trait Implementations

impl Clone for PoolingType

fn clone(&self) -> PoolingType

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for PoolingType

impl Debug for PoolingType

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

Formats the value using the given formatter.

impl Eq for PoolingType

impl PartialEq for PoolingType

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

Equality operator ==.

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

Inequality operator !=.

impl StructuralPartialEq for PoolingType