List your company

Text Generation on 1 Billion Words Dataset (LM1B) (test)

0.71MAUVEAR + DiffusionBlocks
0.48120.54060.60.6594Jun 17, 2025

Evaluation results

MethodLinks
AR + DiffusionBlocksArchitecture=12-layer Llama-2-style, Blocks=B=4, Memory Reduction=4x2025.060.7112.3230.99
Huginn + DiffusionBlockstraining_mode=single-pass, iterations=02025.060.716.0842.43
ARArchitecture=12-layer Llama-2-style2025.060.514.5838.87
Huginntraining_mode=8-step truncated BPTT, iterations=322025.060.4917.0446.73