List your company

Commonsense Reasoning on Hella

49.6AccuracyRoPE++EC
44.19245.5964748.404Dec 8, 2025Dec 11, 2025Dec 15, 2025Dec 18, 2025Dec 22, 2025Dec 25, 2025Dec 29, 2025

Evaluation results

MethodLinks
RoPE++ECModel Scale=1.5B, Context Length Setting=Long2025.1249.6
RoPEModel Scale=1.5B, Context Length Setting=Long2025.1249.3
RoPE++EHModel Scale=1.5B, Context Length Setting=Long2025.1248
RoPE++ECModel Scale=1.5B, Context Length Setting=Short2025.1246.3
RoPEModel Scale=1.5B, Context Length Setting=Short2025.1246.2
RoPE++EHModel Scale=1.5B, Context Length Setting=Short2025.1245.6
weight decayBackbone=Qwen-0.6B2025.1245.4
baselineBackbone=Qwen-0.6B2025.1245.1
NEFTuneBackbone=Qwen-0.6B2025.1244.6
EntroDropBackbone=Qwen-0.6B2025.1244.5
hidden dropoutBackbone=Qwen-0.6B2025.1244.4
vanilla token dropoutBackbone=Qwen-0.6B2025.1244.4