Popular repositories Loading

1. spectral-gap-statement spectral-gap-statement Public

   The Spectral Gap-Statement: when the negative subspace of attention transport is a well-posed invariant

   TeX

2. matched-magnitude-controls matched-magnitude-controls Public

   Matched-magnitude controls for causal testing of spectral mechanisms in attention — a short methodological note

   TeX

3. dcr-attention_v3 dcr-attention_v3 Public

   Top-K sparse attention for long-context decode on Llama-3.2-1B (v3 release)

   Python

4. dcr-attention-paper dcr-attention-paper Public

   Training-free Q-axis sparse attention for long-context LLM decode — exact ranking, a measured entropy-concentration scaling law, long-range retention to 128K, and a per-query accuracy certificate, …

   TeX

5. dcr-retention-no-kcrit dcr-retention-no-kcrit Public

   Top-K sparse attention has no critical key budget: a 4× swing of k_eff barely moves long-context retrieval accuracy across 3 models (Llama-1B/3B, Qwen2.5-3B). The limit is the base model's disambig…

   TeX

6. qaxis-exact-scaling qaxis-exact-scaling Public

   Training-free EXACT-ranking top-K sparse attention. Ranking keys by ⟨u_Q,K⟩ (u_Q=Q/‖Q‖) is order-identical to true attention score. Entropy H_N=α·log N+β (α≈0.31) gives sublinear support N^α and PP…

   TeX