Block-Sparse FlashAttention Perplexity Parity in Multilingual Corpora with Llama-3 Base Models

Abstract

Abstract: Modern large language models increasingly require long contexts for reasoning and multi-document tasks, but attention's quadratic complexity creates a severe computational bottleneck. We present Block-Sparse FlashAttention (BSFA), a drop-in replacement that accelerates long-context inference while preserving model quality. Unlike methods that predict importance before computing scores, BSFA computes exact query-key similarities to select the top-k most important value blocks for each query. By comparing per-block maximum scores against calibrated thresholds, we skip approximately 50\% of the co

Research Question

Does Block-Sparse FlashAttention maintain perplexity parity with full attention on multilingual corpora better than local window attention strategies when evaluated on Llama-3 base models?

Verification Level

Descriptive public verification status only; aggregate claim counts are public, but individual claim records are not exposed here.

Truth-Engine Gate Verdict

This record has passed Gate 2: a Lean4 proof source type-checks, or a sealed-sandbox run reproduced the reported results within the stated tolerance. A reproducible artifact (proof source or repro script and results) is attached to this record. VERIFIED requires an attached reproducible artifact (Lean4 proof source, or repro script and results) before this status can be set; it is not derived from review score or claim count.

Quality Tier

Descriptive public triage only; this tier does not alter current publication or DOI behavior.

Quality Dimensions

Automated triage signals derived from public fields; not human peer review or independent validation.

Correction Record

Public corrections are additive records. Current status does not claim the synthesis is error-free.

Provenance