ADSA halves the KV-cache budget in autoregressive image generation by combining a fixed prefix, a local window, and a diversity-based selection of earlier tokens, with near-identical FID and CLIP scores on LlamaGen.
Autore- gressive model beats diffusion: Llama for scalable image generation.CoRR, 2024
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Make It Efficient: Dynamic Sparse Attention for Autoregressive Image Generation
ADSA halves the KV-cache budget in autoregressive image generation by combining a fixed prefix, a local window, and a diversity-based selection of earlier tokens, with near-identical FID and CLIP scores on LlamaGen.