Struct NetworkHandle 

pub struct NetworkHandle {

    pub name: String,
    pub module: Option<Py<PyAny>>,
    pub optimizer: Option<Py<PyAny>>,
    pub scheduler: Option<Py<PyAny>>,
    pub batching: bool,
    pub k: Option<usize>,
    pub det: bool,
    pub train_mode: bool,
    pub cache_enabled: bool,
    pub cache_size: usize,
    pub arity: Option<usize>,
    pub arg_sorts: Option<Vec<i64>>,
    pub artifact_hash: Option<String>,
}

Handle to a registered neural network.

This struct holds the PyTorch module and associated training state. When the python feature is enabled, it can hold PyO3 Py<PyAny> references.

Fields

name: String

Unique name identifying this network

module: Option<Py<PyAny>>

The PyTorch nn.Module (set via Python API) Only available with the python feature

optimizer: Option<Py<PyAny>>

The optimizer for training (e.g., Adam, SGD) Only available with the python feature

scheduler: Option<Py<PyAny>>

Learning rate scheduler Only available with the python feature

batching: bool

Whether to batch inputs for efficient GPU processing

k: Option<usize>

Top-k sampling: if Some(k), only consider top k outputs

det: bool

Deterministic mode: use argmax instead of sampling

train_mode: bool

Whether the network is in training mode

cache_enabled: bool

Whether output caching is enabled

cache_size: usize

Maximum number of cached outputs

arity: Option<usize>

Registration-time arity, as passed to register_network (see NetworkConfig::arity). Carried through unchanged so it can be read back after registration; the handle does not interpret it.

arg_sorts: Option<Vec<i64>>

Registration-time per-argument catalog sort ids (see NetworkConfig::arg_sorts). Carried through unchanged, like arity.

artifact_hash: Option<String>

Registration-time artifact content hash (see NetworkConfig::artifact_hash). Carried through unchanged, like arity.

Implementations

impl NetworkHandle

pub fn new(name: String) -> Self

Create a new network handle with the given name and default settings.

pub fn from_config(config: &NetworkConfig) -> Self

Create a handle from a configuration.

pub fn has_module(&self) -> bool

Check if the PyTorch module has been set.

pub fn has_optimizer(&self) -> bool

Check if an optimizer has been configured.

pub fn has_scheduler(&self) -> bool

Check if a scheduler has been configured.

pub fn set_module(&mut self, module: Py<PyAny>)

Set the PyTorch module.

pub fn set_optimizer(&mut self, optimizer: Py<PyAny>)

Set the optimizer.

pub fn set_scheduler(&mut self, scheduler: Py<PyAny>)

Set the learning rate scheduler.

pub fn module(&self) -> Option<&Py<PyAny>>

Get a reference to the PyTorch module.

pub fn optimizer(&self) -> Option<&Py<PyAny>>

Get a reference to the optimizer.

pub fn scheduler(&self) -> Option<&Py<PyAny>>

Get a reference to the scheduler.

pub fn clear(&mut self)

Clear the module and training state.

Trait Implementations

impl Debug for NetworkHandle

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

Formats the value using the given formatter.

impl Default for NetworkHandle

fn default() -> Self

Returns the “default value” for a type.