The Condensate Theorem: Transformers are O(n), Not $O(n^2)$
This addresses the scalability problem for large language model inference by eliminating the quadratic bottleneck, potentially reducing costs by >99.9%.
The paper tackles the quadratic computational bottleneck in transformer attention by proving that attention sparsity is a learned topological property, enabling lossless compression to linear complexity with 100% output equivalence, validated by a 159x speedup at 131K tokens and projected >1,200x speedup at 1M tokens.
We present the Condensate Theorem: attention sparsity is a learned topological property, not an architectural constraint. Through empirical analysis of trained language models, we find that attention mass concentrates on a distinct topological manifold -- and this manifold can be identified dynamically without checking every position. We prove a general result: for any query, projecting attention onto the Condensate Manifold (Anchor + Window + Dynamic Top-k) achieves 100% output equivalence with full $O(n^2)$ attention. This is not an approximation -- it is lossless parity. We validate this across GPT-2, Pythia, Qwen2, TinyLlama, and Mistral, demonstrating bit-exact token matching on 1,500+ generated tokens. By mapping this topology to hardware, our Topological Attention kernel achieves a 159x measured speedup at 131K tokens (3.94ms vs 628ms) and a projected >1,200x speedup at 1M tokens, reducing inference costs by >99.9% compared to Flash Attention. We conclude that the quadratic bottleneck is an artifact of naive implementation, not intelligence.