Skip to content
This repository was archived by the owner on Apr 8, 2025. It is now read-only.

Refactor DPR modeling for multigpu training#601

Merged
tholor merged 8 commits intomasterfrom
dpr_multigpu
Oct 26, 2020
Merged

Refactor DPR modeling for multigpu training#601
tholor merged 8 commits intomasterfrom
dpr_multigpu

Conversation

@tholor
Copy link
Copy Markdown
Member

@tholor tholor commented Oct 22, 2020

  • Move calc. of similarity scores from forward() to _embeddings_to_scores()
  • forward() now just passes the query and passage embedding
  • logits_to_loss() converts embeddings => similarity scores => loss

This will simplify later usage of DDP where we need to gather all embeddings for in-batch negative loss

@tholor tholor requested a review from kolk October 23, 2020 10:40
@tholor tholor changed the title WIP Refactor DPR modeling for multigpu training Refactor DPR modeling for multigpu training Oct 23, 2020
@tholor tholor merged commit a3b76b1 into master Oct 26, 2020
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants