|
pulsatrix 1.1.0
|
One collated batch, on the collator's backend. More...
#include <protein_training.hpp>

Public Attributes | |
| Tensor | ids |
(N, L) token ids: masked inputs, padded with <pad>. | |
| Tensor | keep |
(N, L): 1 for a real token, 0 for padding (EncoderLM::set_padding_mask). | |
| Tensor | targets |
(N, L): the original token where one was picked, -100 (ignored) elsewhere. | |
| int64_t | num_targets = 0 |
| Picked tokens: the loss's normalizer. | |
One collated batch, on the collator's backend.
| Tensor pulsatrix::MaskedLMBatch::ids |
(N, L) token ids: masked inputs, padded with <pad>.
| Tensor pulsatrix::MaskedLMBatch::keep |
(N, L): 1 for a real token, 0 for padding (EncoderLM::set_padding_mask).
| int64_t pulsatrix::MaskedLMBatch::num_targets = 0 |
Picked tokens: the loss's normalizer.
| Tensor pulsatrix::MaskedLMBatch::targets |
(N, L): the original token where one was picked, -100 (ignored) elsewhere.