Add content-only token_emb to UnifiedMultimodalEmbeddingModel output#42
Open
Rian354 wants to merge 1 commit into
Open
Add content-only token_emb to UnifiedMultimodalEmbeddingModel output#42Rian354 wants to merge 1 commit into
Rian354 wants to merge 1 commit into
Conversation
forward() now also returns "token_emb": the per-event content embedding before the time and type embeddings are added, in the same temporally-sorted order as "sequence". This is the correct reconstruction target for masked-modeling pretrainers (MAE/SimMIM). The time and type components of "sequence" are largely recoverable from event position, so reconstructing the composed "sequence" dilutes the content learning signal. The change is additive and backward-compatible (a new output key only; all existing keys are unchanged). Adds a test asserting sequence == token_emb + time_embed(time) + type_embedding(type_ids).
|
lgtm! |
will-pang
approved these changes
Jun 24, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
forward() now also returns token_emb, the per-event content embedding before time/type embeddings are added, in the same temporally-sorted order as sequence. This is the right reconstruction target for masked modeling pretrainers (MAE/SimMIM): the time/type components are largely recoverable from event position, so reconstructing the sequence dilutes the content signal.
This PR s purely additive, all existing output keys are unchanged. Includes a test asserting sequence == token_emb + time_embed(time) + type_embedding(type_ids).