BFLA: Block-Filtered Long-Context Attention Mechanism

TL;DR

BFLA introduces a training-free sparse attention mechanism that accelerates long-context inference in large language models with minimal accuracy loss.

Contribution

It presents a novel two-stage block-filtered attention method that can be integrated into existing models without retraining or modification.

Findings

BFLA significantly speeds up long-context prefilling.

Minimal accuracy degradation compared to dense attention methods.

Compatible with multiple large language model series.

Abstract

This paper proposes Block-Filtered Long-Context Attention (BFLA), a training-free sparse prefill attention mechanism for long-context inference. BFLA adopts a two-stage design. In Stage 1, query and key sequences are compressed into coarse blocks, and lightweight block-level softmax mass estimation is performed to construct an input-dependent block importance mask. In Stage 2, the coarse mask is expanded to the Triton attention-tile grid. Several tile-level rescue strategies are applied to reduce information loss, where a fused sparse prefill kernel skips unimportant KV tiles while preserving exact token-level attention inside every retained tile. BFLA requires no retraining, calibration, preprocessing, or model modification and can be plugged into existing vLLM-style paged-attention workloads. Experiments on Gemma 4, Llama 3.1, Qwen 3.5, and Qwen 3.6 series models show that BFLA substantially accelerates long-context prefilling with minimal accuracy degradation compared to dense Triton FlashAttention. Project website: https://github.com/Alicewithrabbit/BFLA.