List your company

Language Modeling on WikiText-2 raw-v1 (test)

2.5Delta PPL (n=1)Interchange (clustered)
1.9325.7669.613.434May 15, 2026

Evaluation results

MethodLinks
Interchange (clustered)Backbone=Qwen3-8B, Precision=bf16, Words=5K, Window=512, Stride=2562026.052.510.345.4222139
SLEB-iterativeBackbone=Qwen3-8B, Precision=bf16, Words=5K, Window=512, Stride=2562026.052.511.246.2———
BI-guidedBackbone=Qwen3-8B, Precision=bf16, Words=5K, Window=512, Stride=2562026.052.521.6137.7———
InterchangeBackbone=Pythia-1.4B, Precision=fp32, Words=5K, Window=512, Stride=2562026.051161.8377.8663616
BI-guidedBackbone=Pythia-1.4B, Precision=fp32, Words=5K, Window=512, Stride=2562026.051174.7186.9———
ReplacementBackbone=Qwen3-8B, Precision=bf16, Words=5K, Window=512, Stride=2562026.0511.348.2116.5———
SLEB-iterativeBackbone=Pythia-1.4B, Precision=fp32, Words=5K, Window=512, Stride=2562026.0511.661.9658.7———
ReplacementBackbone=Pythia-1.4B, Precision=fp32, Words=5K, Window=512, Stride=2562026.0516.71722,310.6———