Struct ModelWeights 

pub struct ModelWeights {
    pub tensors: HashMap<String, Tensor>,
    pub metadata: WeightMetadata,
    pub gguf_config: Option<GGUFModelConfig>,
    pub gguf_tokenizer: Option<GGUFTokenizer>,
    pub quantized_tensors: HashMap<String, Arc<QMatMul>>,
}

Model weights container

Fields

tensors: HashMap<String, Tensor>

Tensor weights by name (F32 dequantized, for compatibility)

metadata: WeightMetadata

Metadata

gguf_config: Option<GGUFModelConfig>

GGUF-specific config (populated when loading from GGUF)

gguf_tokenizer: Option<GGUFTokenizer>

GGUF tokenizer data (populated when loading from GGUF)

quantized_tensors: HashMap<String, Arc<QMatMul>>

Quantized weights (QMatMul) for efficient inference

Implementations

impl ModelWeights

pub fn new(tensors: HashMap<String, Tensor>, metadata: WeightMetadata) -> Self

Create new model weights

pub fn with_gguf_config( tensors: HashMap<String, Tensor>, metadata: WeightMetadata, gguf_config: GGUFModelConfig, ) -> Self

Create new model weights with GGUF config

pub fn with_gguf_config_and_tokenizer( tensors: HashMap<String, Tensor>, metadata: WeightMetadata, gguf_config: GGUFModelConfig, gguf_tokenizer: Option<GGUFTokenizer>, ) -> Self

Create new model weights with GGUF config and tokenizer

pub fn with_quantized( tensors: HashMap<String, Tensor>, metadata: WeightMetadata, gguf_config: GGUFModelConfig, gguf_tokenizer: Option<GGUFTokenizer>, quantized_tensors: HashMap<String, Arc<QMatMul>>, ) -> Self

Create new model weights with GGUF config, tokenizer, and quantized tensors

pub fn get(&self, name: &str) -> Option<&Tensor>

Get tensor by name

pub fn require(&self, name: &str) -> Result<&Tensor>

Get tensor by name (required)

pub fn get_quantized(&self, name: &str) -> Option<Arc<QMatMul>>

Get quantized tensor (QMatMul) by name

pub fn has_quantized(&self, name: &str) -> bool

Check if a tensor has a quantized version

pub fn tensor_names(&self) -> Vec<&String>

Get all tensor names

pub fn to_device(&mut self, device: &Device) -> Result<()>

Move all tensors to device

Trait Implementations

impl Clone for ModelWeights

fn clone(&self) -> ModelWeights

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ModelWeights

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

Formats the value using the given formatter.