pulsatrix 1.1.0
Loading...
Searching...
No Matches
pulsatrix::Transcoder Class Reference

A transcoder: z = ReLU(TopK_k(x W_enc + b_enc)), ŷ = z W_dec + b_dec [+ x W_skip], trained to predict an MLP's output y from its input x by mean((ŷ - y)²) plus AuxK. More...

#include <transcoder.hpp>

Inheritance diagram for pulsatrix::Transcoder:
Collaboration diagram for pulsatrix::Transcoder:

Public Member Functions

 Transcoder (int64_t input_dim, int64_t output_dim, int64_t num_features, DeviceBackend *backend, TranscoderOptions options={})
 
void initialize_bias (const Tensor &target)
 Sets b_dec to the mean of target, (N, output_dim).
 
int64_t input_dim () const override
 
int64_t output_dim () const override
 What decode() produces: input_dim() for an autoencoder.
 
int64_t num_features () const override
 
Tensor encode (const Tensor &x) override
 The codes for a batch, (N, num_features).
 
Tensor decode (const Tensor &codes) override
 codes W_dec + b_dec: the prediction without the skip connection.
 
Tensor predict (const Tensor &x) override
 The full prediction, skip connection included.
 
FeaturizerLoss loss_and_backward (const Tensor &x, std::vector< float > *codes=nullptr) override
 
FeaturizerLoss loss_and_backward_with_target (const Tensor &x, const Tensor &target, std::vector< float > *codes=nullptr) override
 The loss and its gradients; reconstruction is the prediction's mean squared error, sparsity the weighted AuxK loss. Counts the batch toward dead-latent tracking.
 
void normalize_decoder () override
 Rescales every feature so its decoder direction has unit L2 norm, scaling its code by the inverse where the encoder allows it, so reconstructions don't change. Training with unit-norm decoders keeps a sparsity penalty from being dodged by shrinking codes and growing directions, and makes code sizes comparable across features.
 
std::vector< float > decoder_direction (int64_t i) override
 Feature i's decoder direction, output_dim values.
 
Module & parameters_module () override
 The trainable parameters, for an optimizer's step() and zero_grad(), and for checkpoints.
 
std::vector< int64_t > dead_latents () const
 
const std::vector< int64_t > & inputs_since_fired () const
 
void set_inputs_since_fired (const std::vector< int64_t > &since)
 
const TranscoderOptions & options () const
 
LinearModule & encoder ()
 
LinearModule & decoder ()
 
LinearModule * skip ()
 The skip connection; null unless options.skip.
 
Tensor backward (const Tensor &grad_output) override
 Backward through the latents (and the skip connection), adding to the parameters' gradients.
 
Tensor propagate_relevance (const Tensor &relevance_out, const LRPRuleConfig &config) override
 Relevance of the input: the output's relevance is split between the latents' part and the skip's in proportion to their contributions (epsilon rule), then each passes it back with its own layers' rules.
 
OpType op_type () const override
 This module's operation-category tag, for ComputationGraph node tagging.
 
std::vector< NamedParamRef > named_parameters () override
 This module's trainable parameters, each with its hierarchical name – the one place a module declares its parameters (roadmap FND-1).
 
std::optional< DeviceType > compute_device () const override
 The device this module computes on, so forward() can reject an input on another device before any kernel sees it (roadmap FND-8).
 
void set_training (bool training) override
 Sets this module's training/eval mode. Defaults to training (matches every mainstream framework's Module default).
 
void release_activations () override
 Frees the tensors forward() kept for backward() and the relevance passes; containers release their children too. Inference that only needs the output calls it after each forward, so a deep model holds one layer's activations at a time (a 1024-token ESM-2 650M pass otherwise keeps over 10 GB of them).
 
- Public Member Functions inherited from pulsatrix::Module
virtual ~Module ()=default
 
Tensor forward (const Tensor &input)
 Runs this module's forward computation.
 
std::pair< Tensor, NodeId > forward_traced (const Tensor &input, NodeId input_node, ComputationGraph &graph, Autograd &autograd)
 Runs forward() while also registering a ComputationGraph node (tagged with this module's op_type(), parented to input_node) and wiring an Autograd backward function that reuses this module's own backward() – the opt-in traced/explainable path, per Phase 2 Mission 0.
 
virtual bool supports_lrp_rule (LRPRule rule) const
 Whether propagate_relevance() implements rule (no silent fallback: callers such as ExplainerContext::relevance_pass() throw rather than run a module on a rule it does not implement).
 
virtual std::vector< NamedBufferRef > named_buffers ()
 This module's buffers: state a checkpoint must save that no optimizer updates.
 
virtual std::vector< NamedModuleRef > named_children ()
 This module's direct sub-modules, each under the prefix its parameters carry in named_parameters(), in the same order (TRN-7: how ApplyLora() finds a model's Linear layers). Default: none. NamedModules() walks the whole tree.
 
virtual std::vector< ParamRef > parameters ()
 This module's trainable parameters and their gradients, for an optimizer to update uniformly across module types.
 
void set_requires_grad (bool requires_grad, const std::string &prefix="")
 Freezes (false) or unfreezes (true) parameters by name (roadmap FND-2).
 
bool is_training () const
 Whether this module is currently in training mode.
 
- Public Member Functions inherited from pulsatrix::Featurizer
virtual ~Featurizer ()=default
 
virtual int64_t block_size () const
 Codes come in blocks of this many (FEAT-6): a block-sparse featurizer's feature is a block, active when any of its codes is nonzero, and its codes may be negative. 1 (the default) for featurizers whose features are single directions.
 

Protected Member Functions

Tensor forward_impl (const Tensor &input) override
 

Detailed Description

A transcoder: z = ReLU(TopK_k(x W_enc + b_enc)), ŷ = z W_dec + b_dec [+ x W_skip], trained to predict an MLP's output y from its input x by mean((ŷ - y)²) plus AuxK.

  • Why. An SAE explains the residual stream; a transcoder explains a computation. Its latents read from the MLP's input and write to its output, so a circuit through the MLP can be traced latent to latent, and Paulo et al. find its latents more interpretable than an SAE's at the same sparsity.
  • The skip connection takes the MLP's linear part off the latents. decode() leaves it out (it needs the input); predict() includes it.
  • Decoder directions are unit vectors after normalize_decoder(), and the decoder's gradient loses its component along each, as in TopKSparseAutoencoder.
  • Initialization: random unit decoder directions, the encoder their transpose (when the input and output sizes match; random otherwise), b_enc = 0, W_skip = 0, and b_dec 0 until initialize_bias() sets it to the targets' mean.

A Module from input to prediction, with parameters encoder.*, decoder.* and skip.weight. Train it with loss_and_backward_with_target() or TrainFeaturizer(t, x, target, optimizer); loss_and_backward() without a target throws.

Constructor & Destructor Documentation

◆ Transcoder()

pulsatrix::Transcoder::Transcoder ( int64_t  input_dim,
int64_t  output_dim,
int64_t  num_features,
DeviceBackend *  backend,
TranscoderOptions  options = {} 
)
Exceptions
std::invalid_argumentif a size < 1, k outside [1, num_features], k_aux < 0, aux_coefficient < 0 or dead_after < 1.

Member Function Documentation

◆ backward()

Tensor pulsatrix::Transcoder::backward ( const Tensor &  grad_output)
overridevirtual

Backward through the latents (and the skip connection), adding to the parameters' gradients.

Implements pulsatrix::Module.

◆ compute_device()

std::optional< DeviceType > pulsatrix::Transcoder::compute_device ( ) const
inlineoverridevirtual

The device this module computes on, so forward() can reject an input on another device before any kernel sees it (roadmap FND-8).

Returns
std::nullopt (the default) skips the check: a container whose layers check their own inputs, or a user module written before this existed. Every in-tree layer returns its device; EmbeddingModule doesn't, because it reads its indices through their own backend and so accepts them from any device.

Reimplemented from pulsatrix::Module.

◆ dead_latents()

std::vector< int64_t > pulsatrix::Transcoder::dead_latents ( ) const

◆ decode()

Tensor pulsatrix::Transcoder::decode ( const Tensor &  codes)
overridevirtual

codes W_dec + b_dec: the prediction without the skip connection.

Implements pulsatrix::Featurizer.

◆ decoder()

LinearModule & pulsatrix::Transcoder::decoder ( )
inline

◆ decoder_direction()

std::vector< float > pulsatrix::Transcoder::decoder_direction ( int64_t  i)
overridevirtual

Feature i's decoder direction, output_dim values.

Implements pulsatrix::Featurizer.

◆ encode()

Tensor pulsatrix::Transcoder::encode ( const Tensor &  x)
overridevirtual

The codes for a batch, (N, num_features).

Implements pulsatrix::Featurizer.

◆ encoder()

LinearModule & pulsatrix::Transcoder::encoder ( )
inline

◆ forward_impl()

Tensor pulsatrix::Transcoder::forward_impl ( const Tensor &  input)
overrideprotectedvirtual
Exceptions
std::invalid_argumentunless the input is (N, input_dim), N > 0.

Implements pulsatrix::Module.

◆ initialize_bias()

void pulsatrix::Transcoder::initialize_bias ( const Tensor &  target)

Sets b_dec to the mean of target, (N, output_dim).

◆ input_dim()

int64_t pulsatrix::Transcoder::input_dim ( ) const
inlineoverridevirtual

Implements pulsatrix::Featurizer.

◆ inputs_since_fired()

const std::vector< int64_t > & pulsatrix::Transcoder::inputs_since_fired ( ) const
inline

◆ loss_and_backward()

FeaturizerLoss pulsatrix::Transcoder::loss_and_backward ( const Tensor &  x,
std::vector< float > *  codes = nullptr 
)
overridevirtual
Exceptions
std::invalid_argumentalways: a transcoder needs a target.

Implements pulsatrix::Featurizer.

◆ loss_and_backward_with_target()

FeaturizerLoss pulsatrix::Transcoder::loss_and_backward_with_target ( const Tensor &  x,
const Tensor &  target,
std::vector< float > *  codes = nullptr 
)
overridevirtual

The loss and its gradients; reconstruction is the prediction's mean squared error, sparsity the weighted AuxK loss. Counts the batch toward dead-latent tracking.

Reimplemented from pulsatrix::Featurizer.

◆ named_parameters()

std::vector< NamedParamRef > pulsatrix::Transcoder::named_parameters ( )
overridevirtual

This module's trainable parameters, each with its hierarchical name – the one place a module declares its parameters (roadmap FND-1).

Returns
{name, {value, grad}} entries pointing directly at this module's own members, in a fixed order. Names are unique within the module tree. Default: empty (a parameterless module like ReluModule needs no override).
Note
Override this, not parameters(): saving, loading, freezing by name and optimizer parameter groups all key on these names.

Reimplemented from pulsatrix::Module.

◆ normalize_decoder()

void pulsatrix::Transcoder::normalize_decoder ( )
overridevirtual

Rescales every feature so its decoder direction has unit L2 norm, scaling its code by the inverse where the encoder allows it, so reconstructions don't change. Training with unit-norm decoders keeps a sparsity penalty from being dodged by shrinking codes and growing directions, and makes code sizes comparable across features.

Implements pulsatrix::Featurizer.

◆ num_features()

int64_t pulsatrix::Transcoder::num_features ( ) const
inlineoverridevirtual

Implements pulsatrix::Featurizer.

◆ op_type()

OpType pulsatrix::Transcoder::op_type ( ) const
inlineoverridevirtual

This module's operation-category tag, for ComputationGraph node tagging.

Returns
This module's OpType (see op_type.hpp's closed set).
Note
Added in Phase 2 Mission 0 alongside backward() – lets graph-wiring code tag nodes generically through a Module* rather than switching on concrete subclass.

Implements pulsatrix::Module.

◆ options()

const TranscoderOptions & pulsatrix::Transcoder::options ( ) const
inline

◆ output_dim()

int64_t pulsatrix::Transcoder::output_dim ( ) const
inlineoverridevirtual

What decode() produces: input_dim() for an autoencoder.

Reimplemented from pulsatrix::Featurizer.

◆ parameters_module()

Module & pulsatrix::Transcoder::parameters_module ( )
inlineoverridevirtual

The trainable parameters, for an optimizer's step() and zero_grad(), and for checkpoints.

Implements pulsatrix::Featurizer.

◆ predict()

Tensor pulsatrix::Transcoder::predict ( const Tensor &  x)
overridevirtual

The full prediction, skip connection included.

Reimplemented from pulsatrix::Featurizer.

◆ propagate_relevance()

Tensor pulsatrix::Transcoder::propagate_relevance ( const Tensor &  relevance_out,
const LRPRuleConfig &  config 
)
overridevirtual

Relevance of the input: the output's relevance is split between the latents' part and the skip's in proportion to their contributions (epsilon rule), then each passes it back with its own layers' rules.

Implements pulsatrix::Module.

◆ release_activations()

void pulsatrix::Transcoder::release_activations ( )
overridevirtual

Frees the tensors forward() kept for backward() and the relevance passes; containers release their children too. Inference that only needs the output calls it after each forward, so a deep model holds one layer's activations at a time (a 1024-token ESM-2 650M pass otherwise keeps over 10 GB of them).

Note
Until the next forward(), backward() and relevance calls are invalid. The default has nothing to free.

Reimplemented from pulsatrix::Module.

◆ set_inputs_since_fired()

void pulsatrix::Transcoder::set_inputs_since_fired ( const std::vector< int64_t > &  since)
Exceptions
std::invalid_argumentunless it has num_features non-negative values.

◆ set_training()

void pulsatrix::Transcoder::set_training ( bool  training)
overridevirtual

Sets this module's training/eval mode. Defaults to training (matches every mainstream framework's Module default).

Note
Virtual since Phase 6 Mission 6 (SequentialModule) – Mission 5 originally shipped this as plain non-virtual state ("no composite container exists yet to cascade through"); SequentialModule overrides this to cascade to every contained layer, and needs virtual dispatch to do so correctly even when accessed through a Module* base pointer, not just its own concrete type.
BatchNormModule keeps running statistics in training mode and normalizes with them in eval mode; DropoutModule is the identity in eval mode. Call set_training(false) before explaining a model that has either, so a sample's explanation doesn't depend on the rest of its batch.

Reimplemented from pulsatrix::Module.

◆ skip()

LinearModule * pulsatrix::Transcoder::skip ( )
inline

The skip connection; null unless options.skip.


The documentation for this class was generated from the following file: