LongBench Score Discrepancies in Llama-3 Across Evaluation Protocols

Abstract

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: Benchmark archaeology: investigate Longbench score discrepancy for Llama-3 — reported 5.8\%–100.0\% (spread 94.2pp) across 3 papers. Sources: 'Understanding the Performance and Power ' (5.8\%);. 12 claims were extracted from source literature; 2 were independently verified against retrieved documents. An automated multi-reviewer quality assessment produced a score of 5.3/10. This report is a machine-generated literature synthesis and does not constitute original research.

Research Question

Benchmark archaeology: investigate Longbench score discrepancy for Llama-3 — reported 5.8%–100.0% (spread 94.2pp) across 3 papers. Sources: 'Understanding the Performance and Power ' (5.8%); 'SageAttention2: Efficient Attention with' (49.4%); 'ShadowKV: KV Cache in Shadows for High-T' (100.0%). Identify evaluation protocol differences (few-shot, prompting, preprocessing).

Verification Level

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

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