Anchoring Instruction Outside Mask: Exact Reference Caching for Efficient In-Context Diffusion Transformers
Omnimodal generation is central to a wide range of content creation and editing applications. In-context conditioning is essential to this paradigm. It allows diffusion transformers to process text instructions and visual references in a shared attention sequence. However, each reference image introduces thousands of tokens. Computation therefore grows rapidly with the number of references. Existing methods reduce computation through structured sparse attention, which limits interactions between reference and target tokens. This structure also makes the reference K and V independent of the den
Lineage graph
Paper → model → repo connections mined from source citations (Tier-1 exact match).
Why these links exist
Every edge carries a method, confidence, and the source snippet that justified it — so bad links are debuggable.
- FuzzySimilar title/name (fuzzy) · 59%stabilityai/stable-diffusion-xl-base-1.0 →
“Fuzzy title match (0.73): “Anchoring Instruction Outside Mask: Exact Reference Caching ” ≈ “stabilityai/stable-diffusion-xl-base-1.0””
- FuzzySimilar title/name (fuzzy) · 59%CompVis/stable-diffusion-v1-4 →
“Fuzzy title match (0.73): “Anchoring Instruction Outside Mask: Exact Reference Caching ” ≈ “CompVis/stable-diffusion-v1-4””
- FuzzySimilar title/name (fuzzy) · 59%stabilityai/stable-diffusion-3.5-large →
“Fuzzy title match (0.73): “Anchoring Instruction Outside Mask: Exact Reference Caching ” ≈ “stabilityai/stable-diffusion-3.5-large””
- PossiblePossibly related (embedding) · 58%Breakthrough in long-context efficiency announced →
- PossiblePossibly related (embedding) · 55%DiffusionGemma: 4x faster text generation →
- FuzzySimilar title/name (fuzzy) · 84%huggingface/transformers →
“Fuzzy title match (0.92): “Anchoring Instruction Outside Mask: Exact Reference Caching ” ≈ “huggingface/transformers””
- FuzzyOverlapping authors or contributors · 62%modular/modular →
“Shared author/contributor keys: liu”
- FuzzyOverlapping authors or contributors · 62%deepspeedai/DeepSpeed →
“Shared author/contributor keys: lai”