|
pulsatrix 1.1.0
|
Masked-LM quality on held-out sequences. More...
#include <protein_training.hpp>
Public Attributes | |
| double | loss = 0 |
| mean cross-entropy over picked tokens, in nats | |
| double | perplexity = 0 |
| exp(loss) | |
| double | accuracy = 0 |
| share of picked tokens whose argmax is the original | |
| int64_t | tokens = 0 |
Masked-LM quality on held-out sequences.
| double pulsatrix::MaskedLMEvaluation::accuracy = 0 |
share of picked tokens whose argmax is the original
| double pulsatrix::MaskedLMEvaluation::loss = 0 |
mean cross-entropy over picked tokens, in nats
| double pulsatrix::MaskedLMEvaluation::perplexity = 0 |
exp(loss)
| int64_t pulsatrix::MaskedLMEvaluation::tokens = 0 |