Adamas: Hadamard Sparse Attention for Efficient Long-Context Inference

TL;DR

Adamas introduces a novel sparse attention mechanism using Hadamard transforms and efficient selection techniques, enabling accurate long-context inference in language models with significantly reduced computational costs.

Contribution

It proposes a new sparse attention method that maintains accuracy with higher sparsity levels and faster inference for long sequences, outperforming existing approaches.

Findings

Matches full attention accuracy with only 64 tokens budget

Achieves near-lossless performance at 128 tokens

Supports up to 8x higher sparsity with substantial speedups

Abstract

Large language models (LLMs) now support context windows of hundreds of thousands to millions of tokens, enabling applications such as long-document summarization, large-scale code synthesis, multi-document question answering and persistent multi-turn dialogue. However, such extended contexts exacerbate the quadratic cost of self-attention, leading to severe latency in autoregressive decoding. Existing sparse attention methods alleviate these costs but rely on heuristic patterns that struggle to recall critical key-value (KV) pairs for each query, resulting in accuracy degradation. We introduce Adamas, a lightweight yet highly accurate sparse attention mechanism designed for long-context inference. Adamas applies the Hadamard transform, bucketization and 2-bit compression to produce compact representations, and leverages Manhattan-distance estimation for efficient top-k selections. Experiments show that Adamas matches the accuracy of full attention with only a 64-token budget, achieves near-lossless performance at 128, and supports up to 8x higher sparsity than prior state-of-the-art (SOTA) methods while delivering up to 4.4x self-attention and 1.5x end-to-end speedups on 32K-length sequences. Remarkably, Adamas attains comparable or even lower perplexity than full attention, underscoring its effectiveness in maintaining accuracy under aggressive sparsity.