#include <transcoder.hpp>
|
| int64_t | k = 32 |
| | Latents kept per input (a TopK activation, as Paulo et al. use).
|
| |
| bool | skip = false |
| | A skip transcoder: add x W_skip, a linear map from input to output that starts at zero, so the latents only need to explain what isn't linear.
|
| |
| int64_t | k_aux = 0 |
| | AuxK, as TopKSparseAutoencoder: the k_aux largest dead latents predict the error (0: half the output dimension), weighted by aux_coefficient (0 turns it off).
|
| |
| float | aux_coefficient = 1.0f / 32.0f |
| |
| int64_t | dead_after = 10'000'000 |
| |
| uint64_t | seed = 0 |
| |
◆ aux_coefficient
| float pulsatrix::TranscoderOptions::aux_coefficient = 1.0f / 32.0f |
◆ dead_after
| int64_t pulsatrix::TranscoderOptions::dead_after = 10'000'000 |
| int64_t pulsatrix::TranscoderOptions::k = 32 |
Latents kept per input (a TopK activation, as Paulo et al. use).
◆ k_aux
| int64_t pulsatrix::TranscoderOptions::k_aux = 0 |
AuxK, as TopKSparseAutoencoder: the k_aux largest dead latents predict the error (0: half the output dimension), weighted by aux_coefficient (0 turns it off).
◆ seed
| uint64_t pulsatrix::TranscoderOptions::seed = 0 |
◆ skip
| bool pulsatrix::TranscoderOptions::skip = false |
A skip transcoder: add x W_skip, a linear map from input to output that starts at zero, so the latents only need to explain what isn't linear.
The documentation for this struct was generated from the following file: