Struct Param 

pub struct Param<T: Parameter> {
    pub id: ParamId,
    /* private fields */
}

Parameters are the fundamental building blocks of modules where they serve as containers for tensors that can be updated during training, and loaded during inference. If you don’t want to save the tensors and/or don’t want to update it during training, you don’t need this type to wrap your tensor.

Cloning

Cloning a parameter is always cheap; it never allocates or initializes tensors. Clones share the same lazy initialization state, so initialization happens at most once and all clones resolve to the same value regardless of which one triggered it.

This sharing is strictly scoped to lazy initialization. It only guarantees that all clones observe the same initialization result. Subsequent transformations operate on independent parameter values and never propagate across clones.

Fields

id: ParamId

The unique ID of this parameter. This is used by eg. optimizers to associate a gradient with a specific parameter.

Implementations

impl<T: Parameter> Param<T>

pub fn initialized(id: ParamId, value: T) -> Self

Create a new parameter that is already initialized.

pub fn uninitialized<F>( id: ParamId, init: F, device: Device, is_require_grad: bool, shape: Shape, ) -> Self

where F: FnOnce(&Device, bool) -> T + Send + Sync + 'static,

Create a new parameter that is not already initialized.

pub fn val(&self) -> T

Gets the effective parameter value, initializing it lazily if needed.

For initialized parameters, this returns a clone of the cached value. For uninitialized parameters, this triggers initialization.

When a LoRA adapter is attached, this returns the composed value base + scale * (a @ b) rather than the raw stored base. Use base to access the raw stored value without composition.

pub fn base(&self) -> T

Gets the raw stored parameter value (the frozen base when a LoRA adapter is attached), without applying any adapter composition.

pub fn adapter(&self) -> Option<&LoraAdapter>

The LoRA adapter attached to this parameter, if any.

pub fn is_initialized(&self) -> bool

Check if the parameter has been initialized.

Returns true if the parameter’s value has been computed and cached, false if it’s still lazy and will be initialized on first access.

pub fn into_value(self) -> T

Gets the parameter’s value while consuming the parameter.

pub fn consume(self) -> (ParamId, T, ParamMapper<T>)

Gets the parameter id and raw value while consuming the parameter.

Returns the raw stored value (the frozen base when a LoRA adapter is attached); any adapter is dropped. Module traversals strip the adapter before calling into map_float, so mappers always observe the structural base.

pub fn map<F: FnOnce(T) -> T>(self, func: F) -> Self

Execute the given function on the inner value.

pub fn from_mapped_value( id: ParamId, value: T, param_mapper: ParamMapper<T>, ) -> Self

Create an initialized parameter with the given id, value, and param mapper.

This is a helper method for creating parameters while preserving the param mapper, typically used in ModuleMapper implementations.

pub fn load_mapper<F: Fn(T) -> T + Send + Sync + 'static>(self, func: F) -> Self

Runs a transformation on the parameter when loading.

pub fn save_mapper<F: Fn(T) -> T + Send + Sync + 'static>(self, func: F) -> Self

Runs a transformation on the parameter when saving.

pub fn init_mapper<F: Fn(T) -> T + Send + Sync + 'static>(self, func: F) -> Self

where T: Sync + 'static,

Returns a new parameter whose initialization value is transformed by the given function.

If the parameter is still uninitialized (lazy), the transformation is chained onto the existing initialization without triggering evaluation. If the parameter is already initialized, it immediately applies the transformation to the current value.

pub fn lazy_device(&self) -> Device

The device on which the parameter is or will be initialized, without triggering initialization.

This is critical for the load optimization: when loading tensors into an uninitialized parameter, we need to know the target device to move the loaded tensor appropriately, but we don’t want to trigger the initialization function (which would allocate an unnecessary tensor).

Use this instead of crate::tensor::Tensor::device when you need the device but want to preserve lazy initialization.

pub fn set_require_grad(self, require_grad: bool) -> Self

Override the gradient requirement for the current parameter.

pub fn lazy_shape(&self) -> Shape

The shape of the parameter, without triggering initialization.

This is critical for shape validation during loading: when applying tensors to an uninitialized parameter, we need to validate the shape without triggering the initialization function (which would allocate an unnecessary tensor).

Use this instead of crate::tensor::Tensor::shape when you need the shape but want to preserve lazy initialization.

pub fn transform_for_load(self, tensor: T, param_id: ParamId) -> Self

Transform a parameter for loading by applying load transformations.

This method is used to restore a parameter from a tensor (typically during deserialization). It ensures the tensor is moved to the expected device, applies the param mapper’s on_load transformation, and preserves the autodiff settings (require_grad).

pub fn transform_for_save(&self) -> Self

Transform a parameter for saving by applying save transformations.

This method is used to prepare a parameter for saving (typically during serialization). It applies the param mapper’s on_save transformation, which can be used to modify the tensor before serialization (e.g., quantization, precision conversion).

impl<const D: usize> Param<Tensor<D>>

pub fn from_tensor(value: Tensor<D>) -> Self

Create a new parameter from a float tensor.

Warnings

We strongly recommend using Param::uninitialized if you are using this method to initialize parameters inside a module, since the tensor initialization will be lazy, making the loading of weights more performant.

pub fn from_data<T>(data: T, device: &Device) -> Self

where T: Into<TensorData>,

Create a new parameter from data.

Trait Implementations

impl<const D: usize> AutodiffModule for Param<Tensor<D>>

fn valid(&self) -> Self

Returns the same module, but on the inner backend without auto-differentiation.

fn from_inner(module: Self) -> Self

Wraps an inner module back into an auto-diff module.

impl<const D: usize> AutodiffModule for Param<Tensor<D, Int>>

fn valid(&self) -> Self

Returns the same module, but on the inner backend without auto-differentiation.

fn from_inner(module: Self) -> Self

Wraps an inner module back into an auto-diff module.

impl<const D: usize> AutodiffModule for Param<Tensor<D, Bool>>

fn valid(&self) -> Self

Returns the same module, but on the inner backend without auto-differentiation.

fn from_inner(module: Self) -> Self

Wraps an inner module back into an auto-diff module.

impl<T: Parameter> Clone for Param<T>

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<T: Parameter> Debug for Param<T>

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

Formats the value using the given formatter.

impl<T: Parameter> Deref for Param<T>

type Target = T

The resulting type after dereferencing.

fn deref(&self) -> &Self::Target

Dereferences the value.

impl<T: Parameter> Display for Param<T>

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

Formats the value using the given formatter.

impl<const D: usize> Module for Param<Tensor<D>>

fn visit<V: ModuleVisitor>(&self, visitor: &mut V)

Visit each tensor parameter in the module with a visitor.

fn map<M: ModuleMapper>(self, mapper: &mut M) -> Self

Map each tensor parameter in the module with a mapper.

fn to_device(self, device: &Device) -> Self

Move the module and all of its sub-modules to the given device.

fn fork(self, device: &Device) -> Self

Fork the module and all of its sub-modules to the given device.

fn collect_devices(&self, devices: Vec<Device>) -> Vec<Device>

Return all the devices found in the underneath module tree added to the given vector without duplicates.

fn devices(&self) -> Devices

Return all the devices found in the underneath module tree without duplicates.

fn no_grad(self) -> Self

Each tensor in the module tree will not require grad.

fn freeze_group(self, group: ParamGroup) -> Self

Set require_grad to false for every parameter in the given group, leaving the rest of the module untouched.

fn unfreeze_group(self, group: ParamGroup) -> Self

Set require_grad to true for every parameter in the given group, leaving the rest of the module untouched.

fn train(self) -> Self

where Self: AutodiffModule,

Move the module and all of its sub-modules to the autodiff backend.

fn num_params(&self) -> usize

Get the number of parameters the module has, including all of its sub-modules.

fn quantize_weights(self, quantizer: &mut Quantizer) -> Self

Quantize the weights of the module.

fn quantize_weights_group( self, quantizer: &mut Quantizer, group: ParamGroup, ) -> Self

Quantize the weights of the given parameter group.

fn apply_lora(self, config: LoraConfig) -> Self

where Self: Sized,

Attach LoRA adapters to the module’s 2-D weights, freezing the base weights.

fn apply_qlora(self, config: LoraConfig, quantizer: Quantizer) -> Self

where Self: Sized,

Apply QLoRA to the module: quantize the (frozen) base weights and attach trainable LoRA adapters to 2-D weights.

fn into_record(self) -> ModuleRecord

where Self: Sized,

Collect this module’s parameters into a ModuleRecord.

fn try_load_record(self, record: ModuleRecord) -> Result<Self, RecordError>

where Self: Sized,

Apply a ModuleRecord to this module, returning the loaded module.

fn load_record(self, record: ModuleRecord) -> Self

where Self: Sized,

Apply a ModuleRecord to this module, consuming and returning it.

fn save_file<P: AsRef<Path>>(self, path: P) -> Result<(), RecordError>

where Self: Sized,

Available on crate feature std only.

Save this module’s parameters to a burnpack file on disk.

fn load_file<P: AsRef<Path>>(self, path: P) -> Self

where Self: Sized,

Available on crate feature std only.

Load this module’s parameters from a burnpack file on disk, returning the loaded module.

fn try_load_file<P: AsRef<Path>>(self, path: P) -> Result<Self, RecordError>

where Self: Sized,

Available on crate feature std only.

Fallible variant of load_file.

impl<const D: usize> Module for Param<Tensor<D, Int>>

fn visit<V: ModuleVisitor>(&self, visitor: &mut V)

Visit each tensor parameter in the module with a visitor.

fn map<M: ModuleMapper>(self, mapper: &mut M) -> Self

Map each tensor parameter in the module with a mapper.

fn to_device(self, device: &Device) -> Self

Move the module and all of its sub-modules to the given device.

fn fork(self, device: &Device) -> Self

Fork the module and all of its sub-modules to the given device.

fn collect_devices(&self, devices: Vec<Device>) -> Vec<Device>

Return all the devices found in the underneath module tree added to the given vector without duplicates.

fn devices(&self) -> Devices

Return all the devices found in the underneath module tree without duplicates.

fn no_grad(self) -> Self

Each tensor in the module tree will not require grad.

fn freeze_group(self, group: ParamGroup) -> Self

Set require_grad to false for every parameter in the given group, leaving the rest of the module untouched.

fn unfreeze_group(self, group: ParamGroup) -> Self

Set require_grad to true for every parameter in the given group, leaving the rest of the module untouched.

fn train(self) -> Self

where Self: AutodiffModule,

Move the module and all of its sub-modules to the autodiff backend.

fn num_params(&self) -> usize

Get the number of parameters the module has, including all of its sub-modules.

fn quantize_weights(self, quantizer: &mut Quantizer) -> Self

Quantize the weights of the module.

fn quantize_weights_group( self, quantizer: &mut Quantizer, group: ParamGroup, ) -> Self

Quantize the weights of the given parameter group.

fn apply_lora(self, config: LoraConfig) -> Self

where Self: Sized,

Attach LoRA adapters to the module’s 2-D weights, freezing the base weights.

fn apply_qlora(self, config: LoraConfig, quantizer: Quantizer) -> Self

where Self: Sized,

Apply QLoRA to the module: quantize the (frozen) base weights and attach trainable LoRA adapters to 2-D weights.

fn into_record(self) -> ModuleRecord

where Self: Sized,

Collect this module’s parameters into a ModuleRecord.

fn try_load_record(self, record: ModuleRecord) -> Result<Self, RecordError>

where Self: Sized,

Apply a ModuleRecord to this module, returning the loaded module.

fn load_record(self, record: ModuleRecord) -> Self

where Self: Sized,

Apply a ModuleRecord to this module, consuming and returning it.

fn save_file<P: AsRef<Path>>(self, path: P) -> Result<(), RecordError>

where Self: Sized,

Available on crate feature std only.

Save this module’s parameters to a burnpack file on disk.

fn load_file<P: AsRef<Path>>(self, path: P) -> Self

where Self: Sized,

Available on crate feature std only.

Load this module’s parameters from a burnpack file on disk, returning the loaded module.

fn try_load_file<P: AsRef<Path>>(self, path: P) -> Result<Self, RecordError>

where Self: Sized,

Available on crate feature std only.

Fallible variant of load_file.

impl<const D: usize> Module for Param<Tensor<D, Bool>>

fn visit<V: ModuleVisitor>(&self, visitor: &mut V)

Visit each tensor parameter in the module with a visitor.

fn map<M: ModuleMapper>(self, mapper: &mut M) -> Self

Map each tensor parameter in the module with a mapper.

fn to_device(self, device: &Device) -> Self

Move the module and all of its sub-modules to the given device.

fn fork(self, device: &Device) -> Self

Fork the module and all of its sub-modules to the given device.

fn collect_devices(&self, devices: Vec<Device>) -> Vec<Device>

Return all the devices found in the underneath module tree added to the given vector without duplicates.

fn devices(&self) -> Devices

Return all the devices found in the underneath module tree without duplicates.

fn no_grad(self) -> Self

Each tensor in the module tree will not require grad.

fn freeze_group(self, group: ParamGroup) -> Self

Set require_grad to false for every parameter in the given group, leaving the rest of the module untouched.

fn unfreeze_group(self, group: ParamGroup) -> Self

Set require_grad to true for every parameter in the given group, leaving the rest of the module untouched.

fn train(self) -> Self

where Self: AutodiffModule,

Move the module and all of its sub-modules to the autodiff backend.

fn num_params(&self) -> usize

Get the number of parameters the module has, including all of its sub-modules.

fn quantize_weights(self, quantizer: &mut Quantizer) -> Self

Quantize the weights of the module.

fn quantize_weights_group( self, quantizer: &mut Quantizer, group: ParamGroup, ) -> Self

Quantize the weights of the given parameter group.

fn apply_lora(self, config: LoraConfig) -> Self

where Self: Sized,

Attach LoRA adapters to the module’s 2-D weights, freezing the base weights.

fn apply_qlora(self, config: LoraConfig, quantizer: Quantizer) -> Self

where Self: Sized,

Apply QLoRA to the module: quantize the (frozen) base weights and attach trainable LoRA adapters to 2-D weights.

fn into_record(self) -> ModuleRecord

where Self: Sized,

Collect this module’s parameters into a ModuleRecord.

fn try_load_record(self, record: ModuleRecord) -> Result<Self, RecordError>

where Self: Sized,

Apply a ModuleRecord to this module, returning the loaded module.

fn load_record(self, record: ModuleRecord) -> Self

where Self: Sized,

Apply a ModuleRecord to this module, consuming and returning it.

fn save_file<P: AsRef<Path>>(self, path: P) -> Result<(), RecordError>

where Self: Sized,

Available on crate feature std only.

Save this module’s parameters to a burnpack file on disk.

fn load_file<P: AsRef<Path>>(self, path: P) -> Self

where Self: Sized,

Available on crate feature std only.

Load this module’s parameters from a burnpack file on disk, returning the loaded module.

fn try_load_file<P: AsRef<Path>>(self, path: P) -> Result<Self, RecordError>

where Self: Sized,

Available on crate feature std only.

Fallible variant of load_file.

impl<const D: usize> ModuleDisplay for Param<Tensor<D>>

fn format(&self, passed_settings: DisplaySettings) -> String

Formats the module with provided display settings.

fn custom_settings(&self) -> Option<DisplaySettings>

Custom display settings for the module.

fn custom_content(&self, _content: Content) -> Option<Content>

Custom attributes for the module.

impl<const D: usize> ModuleDisplay for Param<Tensor<D, Int>>

fn format(&self, passed_settings: DisplaySettings) -> String

Formats the module with provided display settings.

fn custom_settings(&self) -> Option<DisplaySettings>

Custom display settings for the module.

fn custom_content(&self, _content: Content) -> Option<Content>

Custom attributes for the module.

impl<const D: usize> ModuleDisplay for Param<Tensor<D, Bool>>

fn format(&self, passed_settings: DisplaySettings) -> String

Formats the module with provided display settings.

fn custom_settings(&self) -> Option<DisplaySettings>

Custom display settings for the module.

fn custom_content(&self, _content: Content) -> Option<Content>

Custom attributes for the module.

impl<const D: usize> ModuleDisplayDefault for Param<Tensor<D>>

fn content(&self, content: Content) -> Option<Content>

Attributes of the module used for display purposes.

fn num_params(&self) -> usize

Gets the number of the parameters of the module.

impl<const D: usize> ModuleDisplayDefault for Param<Tensor<D, Int>>

fn content(&self, content: Content) -> Option<Content>

Attributes of the module used for display purposes.

fn num_params(&self) -> usize

Gets the number of the parameters of the module.

impl<const D: usize> ModuleDisplayDefault for Param<Tensor<D, Bool>>

fn content(&self, content: Content) -> Option<Content>

Attributes of the module used for display purposes.

fn num_params(&self) -> usize

Gets the number of the parameters of the module.