Fix swapped tensor order in TransformerBlockSkipHook for SkipLayerGuidance#7
Draft
cursor[bot] wants to merge 1 commit into
Draft
Fix swapped tensor order in TransformerBlockSkipHook for SkipLayerGuidance#7cursor[bot] wants to merge 1 commit into
cursor[bot] wants to merge 1 commit into
Conversation
When skipping dual-stream transformer blocks (JointTransformerBlock, FluxTransformerBlock, QwenImageTransformerBlock), the hook returned (hidden_states, encoder_hidden_states) regardless of each block's actual return tuple order. Callers unpack as (encoder_hidden_states, hidden_states), so SkipLayerGuidance silently swapped image and text tensors on skipped layers. Build the skip output using TransformerBlockRegistry indices, matching MagCacheBlockHook. Co-authored-by: Simon Lynch <srlynch1@users.noreply.github.com>
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Bug and impact
TransformerBlockSkipHook(used bySkipLayerGuidance,PerturbedAttentionGuidance, andAutoGuidance) silently corrupted inference for dual-stream transformers such as SD3, Flux, and QwenImage.When a layer is fully skipped (
skip_attention=True,skip_ff=True,dropout=1.0), the hook returned(hidden_states, encoder_hidden_states)unconditionally. Blocks likeJointTransformerBlockandFluxTransformerBlockactually return(encoder_hidden_states, hidden_states). Callers unpack in that order, so every subsequent block received swapped image/text tensors with no exception.Root cause
The hook ignored
TransformerBlockRegistryreturn indices (return_hidden_states_index,return_encoder_hidden_states_index) and always built the tuple in the wrong order for blocks wherehidden_statesis at index 1.Fix and validation
MagCacheBlockHook).tests/hooks/test_layer_skip.pywith a Flux-like block registered at index 1/0; verifies a skipped first block preserves correct(encoder_hidden_states, hidden_states)order through the forward loop.