Struct ModelConfig 

pub struct ModelConfig {
    pub model_info: ModelInfo,
    pub shapes: ShapeConfig,
    pub components: HashMap<String, ComponentConfig>,
    pub naming: NamingConfig,
    pub ffn_execution: Option<String>,
}

Complete model configuration including shapes, components, and naming patterns

Fields

model_info: ModelInfo

shapes: ShapeConfig

components: HashMap<String, ComponentConfig>

naming: NamingConfig

ffn_execution: Option<String>

Execution mode for FFN: “unified” (single component/function) or “split” (separate prefill/infer components)

Implementations

impl ModelConfig

pub fn default_qwen() -> Self

Create a minimal default Qwen ModelConfig (no components). Useful for tests and fallbacks.

pub fn load_from_file<P: AsRef<Path>>(path: P) -> Result<Self>

Load configuration from a JSON file

pub fn save_to_file<P: AsRef<Path>>(&self, path: P) -> Result<()>

Save configuration to a JSON file

pub fn get_tensor_shape( &self, component: &str, tensor_name: &str, is_input: bool, ) -> Option<&Vec<usize>>

Get the shape configuration for a specific component and tensor

pub fn embeddings_input_shape(&self) -> Option<&Vec<usize>>

Get the expected input shape for embeddings

pub fn embeddings_output_shape(&self) -> Option<&Vec<usize>>

Get the expected output shape for embeddings

pub fn ffn_prefill_input_shape(&self) -> Option<&Vec<usize>>

Get the expected input shape for FFN prefill

pub fn lm_head_input_shape(&self) -> Option<&Vec<usize>>

Get the expected input shape for LM head

pub fn has_multipart_logits(&self) -> bool

Check if this model supports multi-part logits output

pub fn logits_part_count(&self) -> usize

Get the number of logits parts for multi-part output

pub fn lm_head_primary_output_name(&self) -> Option<String>

Select the primary logits output name for the LM head. Preference order: “logits1” (multipart), then “logits”, otherwise the first available key.

pub fn validate(&self) -> Result<()>

Validate the configuration for consistency

pub fn validate_internal_wiring(&self) -> Result<()>

Validate internal wiring between components for basic shape compatibility. Examples:

• embeddings.outputs.hidden_states == ffn_prefill.inputs.hidden_states
• ffn_infer.outputs.output_hidden_states == lm_head.inputs.hidden_states (when ffn_infer exists)

pub fn ffn_is_split(&self) -> bool

Determine if FFN execution should be treated as split (separate infer component)

pub fn prefill_is_single_token(&self) -> bool

Detect if prefill should run in single-token sequential mode based on configured shapes

pub fn expects_full_sequence_prefill(&self) -> bool

Check if the model expects full sequence prefill (as opposed to single-token processing) This is typically true for CoreML models with fixed-shape inputs like [1, 128, 1024]

pub fn create_embeddings_input_tensor( &self, tokens: &[i64], device: &Device, ) -> Result<Tensor, CandleError>

Create embeddings input tensor with proper shape from configuration

pub fn create_ffn_position_ids_tensor( &self, positions: &[i64], device: &Device, ) -> Result<Tensor, CandleError>

Create position IDs tensor for FFN prefill with proper shape

pub fn create_ffn_causal_mask_tensor( &self, _batch_size: usize, _context_length: usize, device: &Device, ) -> Result<Tensor, CandleError>

Create causal mask tensor for FFN with proper shape

pub fn create_single_token_hidden_states( &self, _tokens: &[i64], device: &Device, ) -> Result<Tensor, CandleError>

Create single token hidden states tensor for LM head

pub fn create_infer_position_ids_tensor( &self, position: i64, device: &Device, ) -> Result<Tensor, CandleError>

Create position IDs tensor for inference (single position)

pub fn create_current_pos_tensor( &self, position: i64, device: &Device, ) -> Result<Tensor, CandleError>

Create current position tensor for FFN

Trait Implementations

impl Clone for ModelConfig

fn clone(&self) -> ModelConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ModelConfig

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for ModelConfig

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for ModelConfig

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

where __S: Serializer,

Serialize this value into the given Serde serializer.