List your company

Language Modeling Statistics on WikiText-2

93.039Average KurtosisVanilla
-1.5292423.0221347.573572.12487May 18, 2026

Evaluation results

MethodLinks
VanillaModel=Qwen3-0.6B, Precision=FP162026.0593.03974.77810.568
VanillaModel=Qwen3-0.6B, Precision=W8A82026.0593.03974.77836.642
VanillaModel=Qwen3-0.6B, Precision=OmniQuant W6A62026.0593.03974.77844.778
VanillaModel=Qwen3-0.6B, Precision=OmniQuant W4A42026.0593.03974.77850.821
VanillaModel=Llama-3.2-1B, Precision=FP162026.0591.70684.4338.981
VanillaModel=Llama-3.2-1B, Precision=W8A82026.0591.70684.43330.189
VanillaModel=Llama-3.2-1B, Precision=OmniQuant W6A62026.0591.70684.43339.378
VanillaModel=Llama-3.2-1B, Precision=OmniQuant W4A42026.0591.70684.43374.967
Gated AttentionModel=Llama-3.2-1B, Precision=FP162026.054.39414.8898.259
Gated AttentionModel=Llama-3.2-1B, Precision=W8A82026.054.39414.8899.214
Gated AttentionModel=Llama-3.2-1B, Precision=OmniQuant W6A62026.054.39414.88913.103
Gated AttentionModel=Llama-3.2-1B, Precision=OmniQuant W4A42026.054.39414.88924.27
Clipped SoftmaxModel=Llama-3.2-1B, Precision=FP162026.053.29714.2439.049
Clipped SoftmaxModel=Llama-3.2-1B, Precision=W8A82026.053.29714.24310.264
Clipped SoftmaxModel=Llama-3.2-1B, Precision=OmniQuant W6A62026.053.29714.24314.368
Clipped SoftmaxModel=Llama-3.2-1B, Precision=OmniQuant W4A42026.053.29714.24326.808
OutEffHopModel=Llama-3.2-1B, Precision=FP162026.053.07613.7769.07
OutEffHopModel=Llama-3.2-1B, Precision=W8A82026.053.07613.77610.419
OutEffHopModel=Llama-3.2-1B, Precision=OmniQuant W6A62026.053.07613.77613.555
OutEffHopModel=Llama-3.2-1B, Precision=OmniQuant W4A42026.053.07613.77626.183
OutEffHopModel=Qwen3-0.6B, Precision=FP162026.053.03514.1649.991
OutEffHopModel=Qwen3-0.6B, Precision=W8A82026.053.03514.16410.599
OutEffHopModel=Qwen3-0.6B, Precision=OmniQuant W6A62026.053.03514.16413.357
OutEffHopModel=Qwen3-0.6B, Precision=OmniQuant W4A42026.053.03514.16432.362
Gated AttentionModel=Qwen3-0.6B, Precision=FP162026.053.03314.3610.556
Gated AttentionModel=Qwen3-0.6B, Precision=W8A82026.053.03314.3619.261
Gated AttentionModel=Qwen3-0.6B, Precision=OmniQuant W6A62026.053.03314.3623.427
Gated AttentionModel=Qwen3-0.6B, Precision=OmniQuant W4A42026.053.03314.3628.775
AoSModel=Llama-3.2-1B, Precision=FP162026.053.03112.5389
AoSModel=Llama-3.2-1B, Precision=W8A82026.053.03112.5389.364
AoSModel=Llama-3.2-1B, Precision=OmniQuant W6A62026.053.03112.53811.284
AoSModel=Llama-3.2-1B, Precision=OmniQuant W4A42026.053.03112.53822.633
AoSModel=Qwen3-0.6B, Precision=FP162026.053.03113.8989.345
AoSModel=Qwen3-0.6B, Precision=W8A82026.053.03113.8989.615
AoSModel=Qwen3-0.6B, Precision=OmniQuant W6A62026.053.03113.89810.484
AoSModel=Qwen3-0.6B, Precision=OmniQuant W4A42026.053.03113.89817.516
Gated OutEffHopModel=Qwen3-0.6B, Precision=FP162026.053.02913.819.999
Gated OutEffHopModel=Qwen3-0.6B, Precision=W8A82026.053.02913.8110.572
Gated OutEffHopModel=Qwen3-0.6B, Precision=OmniQuant W6A62026.053.02913.8111.521
Gated OutEffHopModel=Qwen3-0.6B, Precision=OmniQuant W4A42026.053.02913.8121.258
Clipped SoftmaxModel=Qwen3-0.6B, Precision=FP162026.053.02714.16410.366
Clipped SoftmaxModel=Qwen3-0.6B, Precision=W8A82026.053.02714.16421.382
Clipped SoftmaxModel=Qwen3-0.6B, Precision=OmniQuant W6A62026.053.02714.16428.736
Clipped SoftmaxModel=Qwen3-0.6B, Precision=OmniQuant W4A42026.053.02714.16439.891
Gated AoSModel=Qwen3-0.6B, Precision=FP162026.053.02513.559.353
Gated AoSModel=Qwen3-0.6B, Precision=W8A82026.053.02513.559.591
Gated AoSModel=Qwen3-0.6B, Precision=OmniQuant W6A62026.053.02513.5510.455
Gated AoSModel=Qwen3-0.6B, Precision=OmniQuant W4A42026.053.02513.5515.432
Clipped OutEffHopModel=Qwen3-0.6B, Precision=FP162026.053.02312.6859.869
Clipped OutEffHopModel=Qwen3-0.6B, Precision=W8A82026.053.02312.68510.437
Clipped OutEffHopModel=Qwen3-0.6B, Precision=OmniQuant W6A62026.053.02312.68511.385
Clipped OutEffHopModel=Qwen3-0.6B, Precision=OmniQuant W4A42026.053.02312.68530.742
Clipped AoSModel=Qwen3-0.6B, Precision=FP162026.053.01912.459.231
Clipped AoSModel=Qwen3-0.6B, Precision=W8A82026.053.01912.459.469
Clipped AoSModel=Qwen3-0.6B, Precision=OmniQuant W6A62026.053.01912.4510.326
Clipped AoSModel=Qwen3-0.6B, Precision=OmniQuant W4A42026.053.01912.4516.949
Clipped OutEffHopModel=Llama-3.2-1B, Precision=FP162026.052.39910.8489.08
Clipped OutEffHopModel=Llama-3.2-1B, Precision=W8A82026.052.39910.84810.188
Clipped OutEffHopModel=Llama-3.2-1B, Precision=OmniQuant W6A62026.052.39910.84813.277
Clipped OutEffHopModel=Llama-3.2-1B, Precision=OmniQuant W4A42026.052.39910.84825.624
Clipped AoSModel=Llama-3.2-1B, Precision=FP162026.052.3649.8749.01
Clipped AoSModel=Llama-3.2-1B, Precision=W8A82026.052.3649.8749.336
Clipped AoSModel=Llama-3.2-1B, Precision=OmniQuant W6A62026.052.3649.87411.25
Clipped AoSModel=Llama-3.2-1B, Precision=OmniQuant W4A42026.052.3649.87422.565
Gated OutEffHopModel=Llama-3.2-1B, Precision=FP162026.052.1398.3258.751
Gated OutEffHopModel=Llama-3.2-1B, Precision=W8A82026.052.1398.3259.198
Gated OutEffHopModel=Llama-3.2-1B, Precision=OmniQuant W6A62026.052.1398.32512.083
Gated OutEffHopModel=Llama-3.2-1B, Precision=OmniQuant W4A42026.052.1398.32523.231
Gated AoSModel=Llama-3.2-1B, Precision=FP162026.052.1088.0268.683
Gated AoSModel=Llama-3.2-1B, Precision=W8A82026.052.1088.0268.875
Gated AoSModel=Llama-3.2-1B, Precision=OmniQuant W6A62026.052.1088.02610.695
Gated AoSModel=Llama-3.2-1B, Precision=OmniQuant W4A42026.052.1088.02621.451