Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Update vilbert.py #1065

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Update vilbert.py #1065

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
6043e09
Update vilbert.py
dk25021999 Aug 31, 2021
2180572
Update vilbert.py
dk25021999 Sep 1, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 13 additions & 0 deletions mmf/models/vilbert.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,7 @@
import torch.nn.functional as F
from mmf.common.registry import registry
from mmf.models import BaseModel
from mmf.models.transformers.heads.itm import ITM
from mmf.modules.hf_layers import replace_with_jit
from mmf.utils.configuration import get_mmf_cache_dir
from mmf.utils.modeling import get_optimizer_parameters_for_bert
Expand Down Expand Up @@ -1061,6 +1062,9 @@ def __init__(self, config):
self.visual_target = config.visual_target
self.num_negative = config.num_negative
self.loss_fct = CrossEntropyLoss(ignore_index=-1)

if itm_loss is not False:
itm_head = ITM({"type": "itm", "hidden_size": self.vocab_size})

if self.visual_target == 0:
self.vis_criterion = nn.KLDivLoss(reduction="none")
Expand Down Expand Up @@ -1099,6 +1103,8 @@ def forward(
image_label: Optional[Tensor] = None,
image_target: Optional[Tensor] = None,
output_all_attention_masks: bool = False,
itm_loss: bool = False,
next_sentence_label: Optional[Dict[str, Dict[str, torch.Tensor]]] = None,
) -> Dict[str, Tensor]:
masked_img_loss: Optional[Tensor] = None
(
Expand Down Expand Up @@ -1226,6 +1232,13 @@ def forward(
prediction_scores_t.view(-1, self.vocab_size), masked_lm_labels.view(-1)
)
output["masked_lm_loss"] = masked_lm_loss.unsqueeze(0)

if itm_loss is not False:
seq_output = torch.cat(sequence_output_t, sequence_output_v)
multimodal_alignment_loss = itm_head(seq_output, processed_sample_list = next_sentence_label)
if multimodal_alignment_loss is not None:
output["itm_loss"] = multimodal_alignment_loss["losses"]["itm_loss"]

# next_sentence_loss = self.loss_fct(
# seq_relationship_score.view(-1, 2), next_sentence_label.view(-1)
# )
Expand Down