DiT now supports sequence conditions. #923
Merged
+113
−67
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Outdated suggestions cannot be applied.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.
When using seq2seq models with DiT, the condition may have the same sequence length as the input.
For example:
[batch, seq_len, dim]
[batch, seq_len, cond_dim]
AdaptiveLayerNormModulation now supports conditions in both
[batch, cond_dim]
and[batch, seq_len, cond_dim]
formats. It outputs conditions in the shape[batch, 1|seq_len, cond_dim]
, depending on whetherseq_len
is present.Accordingly, DiT has been updated to handle rank-3 conditions. The codebase has also become simpler. Previously,
jnp.expand_dims
was scattered across many places, but nowAdaptiveLayerNormModulation
adjusts the rank of the condition to match the input and returns it accordingly.Speech detokenizer will use this DiT.