Where the relevance of the explained tokens went. Every vector has one value per position. The ledger balances exactly: value = prompt + mask + leak (+ relayed in CommitmentMode::StepLocal, where it is an endpoint).
More...
#include <masked_diffusion_lrp.hpp>
|
| CommitmentMode | mode = CommitmentMode::Graph |
| |
| double | value = 0 |
| | The explained value: the targets' logits, each at the step that wrote it, summed. The relevance seeded, as LXT seeds a logit.
|
| |
| std::vector< float > | prompt |
| | Relevance on the prompt (positions not masked in the input), from every step. Zero elsewhere.
|
| |
| std::vector< float > | relayed |
| | Relevance that reached each generated position from later steps. Graph: relayed onward from the step that wrote it. StepLocal: stopped there.
|
| |
| std::vector< float > | mask |
| | Relevance each position absorbed while it was masked, summed over steps: the mask sink.
|
| |
| std::vector< CommitmentHop > | hops |
| | One entry per step explained, in the order explained (the last step first). Steps nothing was seeded at aren't run.
|
| |
| std::vector< std::vector< float > > | step_relevance |
| | For each of the trajectory's steps, the relevance on its input (L values), or empty for a step that wasn't run.
|
| |
Where the relevance of the explained tokens went. Every vector has one value per position. The ledger balances exactly: value = prompt + mask + leak (+ relayed in CommitmentMode::StepLocal, where it is an endpoint).
◆ leak_total()
| double pulsatrix::CommitmentRelevance::leak_total |
( |
| ) |
const |
◆ mask_total()
| double pulsatrix::CommitmentRelevance::mask_total |
( |
| ) |
const |
◆ prompt_total()
| double pulsatrix::CommitmentRelevance::prompt_total |
( |
| ) |
const |
◆ relayed_total()
| double pulsatrix::CommitmentRelevance::relayed_total |
( |
| ) |
const |
◆ hops
One entry per step explained, in the order explained (the last step first). Steps nothing was seeded at aren't run.
◆ mask
| std::vector<float> pulsatrix::CommitmentRelevance::mask |
Relevance each position absorbed while it was masked, summed over steps: the mask sink.
◆ mode
◆ prompt
| std::vector<float> pulsatrix::CommitmentRelevance::prompt |
Relevance on the prompt (positions not masked in the input), from every step. Zero elsewhere.
◆ relayed
| std::vector<float> pulsatrix::CommitmentRelevance::relayed |
Relevance that reached each generated position from later steps. Graph: relayed onward from the step that wrote it. StepLocal: stopped there.
◆ step_relevance
| std::vector<std::vector<float> > pulsatrix::CommitmentRelevance::step_relevance |
For each of the trajectory's steps, the relevance on its input (L values), or empty for a step that wasn't run.
◆ value
| double pulsatrix::CommitmentRelevance::value = 0 |
The explained value: the targets' logits, each at the step that wrote it, summed. The relevance seeded, as LXT seeds a logit.
The documentation for this struct was generated from the following file: