List your company

Language Modeling on LAMBADA (PPL)

4.11Perplexity (PPL)K-stage RK
0.934422.369743.80565.2403May 22, 2026May 23, 2026May 25, 2026May 27, 2026May 29, 2026May 31, 2026Jun 2, 2026

Evaluation results

MethodLinks
K-stage RKSolver choice=Runge–Kutta, K=22026.054.11
Baseline2026.054.12
Heun K=1Solver choice=Heun, K=12026.054.12
MoE Transformer w/ dynamic conv. (low-rank)Train Tokens=100B, Params=6.80B, Conv. Location=QKV2026.0610.92
MoE Transformer w/ static conv.Train Tokens=100B, Params=6.77B, Conv. Location=QKV2026.0611.2
MoE TransformerTrain Tokens=100B, Params=6.77B, Conv. Location=–2026.0611.55
MoE Transformer w/ dynamic conv. (head-wise)Train Tokens=100B, Params=6.80B, Conv. Location=QKV2026.0611.61
Transformer w/ dynamic conv. (low-rank)Train Tokens=100B, Params=1.88B, Conv. Location=all linear2026.0612.51
Transformer w/ dynamic conv. (low-rank)Train Tokens=100B, Params=1.88B, Conv. Location=QKV2026.0615.43
Transformer w/ static conv.Train Tokens=100B, Params=1.83B, Conv. Location=QKV2026.0616.21
Transformer w/ dynamic conv. (head-wise)Train Tokens=100B, Params=1.88B, Conv. Location=QKV2026.0616.34
TransformerTrain Tokens=100B, Params=1.82B, Conv. Location=–2026.0617.28
Transformer w/ more params (wider FFN)Train Tokens=100B, Params=1.87B, Conv. Location=–2026.0618.49
Transformer w/ dynamic conv. (low-rank)Train Tokens=15B, Params=319.0M, Conv. Location=all linear2026.0646.13
Gated DeltaNet w/ dynamic conv. (low-rank)Train Tokens=15B, Params=309.5M, Conv. Location=QKV2026.0650.56
Transformer w/ dynamic conv. (low-rank)Train Tokens=15B, Params=311.8M, Conv. Location=QKV2026.0656.66
Transformer w/ dynamic conv. (head-wise)Train Tokens=15B, Params=311.7M, Conv. Location=QKV2026.0658.54
Gated DeltaNet w/ dynamic conv. (head-wise)Train Tokens=15B, Params=309.6M, Conv. Location=QKV2026.0659.49
Mamba-2 w/ dynamic conv. (head-wise)Train Tokens=15B, Params=309.8M, Conv. Location=QKV2026.0665.8
Gated DeltaNet w/ static conv.Train Tokens=15B, Params=305.4M, Conv. Location=QKV2026.0667.63
Transformer w/ more params (wider FFN)Train Tokens=15B, Params=311.5M, Conv. Location=–2026.0668.05
Transformer w/ static conv.Train Tokens=15B, Params=305.4M, Conv. Location=QKV2026.0669.64
Gated DeltaNet (w/o conv.)Train Tokens=15B, Params=305.2M, Conv. Location=–2026.0669.9
Mamba-2 w/ dynamic conv. (low-rank)Train Tokens=15B, Params=309.8M, Conv. Location=QKV2026.0671.12
TransformerTrain Tokens=15B, Params=305.2M, Conv. Location=–2026.0676.62
Mamba-2 (w/o conv.)Train Tokens=15B, Params=306.2M, Conv. Location=–2026.0680.81
Mamba-2 w/ static conv.Train Tokens=15B, Params=306.4M, Conv. Location=QKV2026.0683.5