Struct DistributedLayeredModel 

pub struct DistributedLayeredModel<M: LayerParallelism> { /* private fields */ }

A LayerParallelism model split across devices, holding where each of its layers runs so the forward reads it rather than working it out again on every call. It is itself a Module, so records, devices and training flags work as they do on the model alone.

Implementations

impl<M: LayerParallelism> DistributedLayeredModel<M>

pub fn new(model: M, placement: &LayerPlacement) -> Self

Split model across the devices placement gives its layers. Nothing moves: each layer is built where it runs, and this checks it is.

Panics

Panics when placement does not give a device to every hidden layer, or when a layer holds a tensor off the device placement gives it.

pub fn placement(&self) -> &LayerPlacement

Where each layer runs, which is where its own parameters are.

pub fn forward( &self, input: <M::InputLayer as DistributedLayer>::Input, ) -> <M::OutputLayer as DistributedLayer>::Output

Run each layer on its device, moving what passes between them.

pub fn model(&self) -> &M

The model, its layers where the placement puts them.

pub fn into_inner(self) -> M

The model on its own, no longer carrying where its layers run.

pub fn into_stages(self) -> StagedLayers<M>

The model taken apart, its hidden layers cut into the placement’s stages.

Panics

Panics when a stage would hold a layer off its device, which is what an into_layers giving the hidden layers in another order than layers_hidden does.

Trait Implementations

impl<M: Clone + LayerParallelism> Clone for DistributedLayeredModel<M>

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<M: Debug + LayerParallelism> Debug for DistributedLayeredModel<M>

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

Formats the value using the given formatter.

impl<M: LayerParallelism> Module for DistributedLayeredModel<M>

A mapper is trusted to leave each parameter where it found it, so only a move of the whole model rewrites the placement.

fn collect_devices(&self, devices: Devices) -> Devices

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

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

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

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

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

fn train(self) -> Self

Enables autodiff and restores configured training state throughout the module.

fn valid(&self) -> Self

Returns a validation snapshot without autodiff or active training flags.

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

Visit each parameter and module-owned control value with a visitor.

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

Map each parameter and module-owned control value with a mapper.

fn materialize(self) -> Self

Fold parameter reparameterizations into their effective values and remove them.

fn devices(&self) -> Devices

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

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

Set whether every floating-point tensor parameter in the module tree requires gradients.

fn set_require_grad_group(self, group: ParamGroup, require_grad: bool) -> Self

Set whether matched floating-point tensor parameters require gradients.

fn no_grad(self) -> Self

Disable gradients for every floating-point tensor parameter in the module tree.

fn freeze(self) -> Self

Freeze the whole module tree.

fn unfreeze(self) -> Self

Unfreeze the whole module tree.

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

Freeze every module-owned value in the given group, leaving the rest of the module untouched.

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

Unfreeze every module-owned value in the given group, leaving the rest of the module untouched.

fn num_params(&self) -> usize

Get the number of parameters the module has, including all of its sub-modules, without initializing the ones not initialized yet.

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_reparameterization<R>(self, reparameterizer: R) -> Self

where Self: Sized, R: Reparameterizer,

Attach reparameterizations using the given Reparameterizer.

fn apply_lora(self, lora: Lora) -> Self

where Self: Sized,

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

fn apply_qlora(self, qlora: QLora) -> Self

where Self: Sized,

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

fn into_record(self) -> ModuleRecord

where Self: Sized,

Collect this module’s tensor parameters into a ModuleRecord.

fn into_record_group(self, group: ParamGroup) -> ModuleRecord

where Self: Sized,

Collect the tensor parameters group names 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.