List your company

Language Modeling Evaluation

1.71Perplexity (PPL)Gated Subspace Inference
0.59068.1465515.702523.25845May 4, 2026May 8, 2026May 13, 2026May 18, 2026May 22, 2026May 27, 2026Jun 1, 2026

Evaluation results

MethodLinks
Gated Subspace InferenceModel=GPT-2 124M (d = 768, L = 12), k/ε=256/0.102026.051.710.986991003
Gated Subspace InferenceModel=GPT-2 124M (d = 768, L = 12), k/ε=128/0.052026.051.720.99599.441.11.5
Gated Subspace InferenceModel=GPT-2 124M (d = 768, L = 12), k/ε=256/0.052026.051.720.99199.491.42.6
Gated Subspace InferenceModel=GPT-J 6B (d = 4096, L = 28), k/ε=256/0.102026.051.720.99198.699.815.6
Gated Subspace InferenceModel=GPT-J 6B (d = 4096, L = 28), k/ε=256/0.152026.051.720.99298.610016
Baseline GPT-2 124MModel=GPT-2 124M (d = 768, L = 12)2026.051.73————
Baseline GPT-J 6BModel=GPT-J 6B (d = 4096, L = 28)2026.051.73————
Gated Subspace InferenceModel=GPT-J 6B (d = 4096, L = 28), k/ε=128/0.052026.051.730.99999.421.11.3
Gated Subspace InferenceModel=GPT-J 6B (d = 4096, L = 28), k/ε=256/0.052026.051.730.99799.277.23.6
Gated Subspace InferenceModel=GPT-J 6B (d = 4096, L = 28), k/ε=128/0.102026.051.741.0029963.62.6
Gated Subspace InferenceModel=GPT-J 6B (d = 4096, L = 28), k/ε=128/0.152026.051.741.00596.581.94.8
Gated Subspace InferenceModel=OPT 6.7B (d = 4096, L = 32), k/ε=256/0.152026.051.80.98699.297.311.4
Gated Subspace InferenceModel=OPT 6.7B (d = 4096, L = 32), k/ε=128/0.052026.051.810.98998.245.71.8
Gated Subspace InferenceModel=OPT 6.7B (d = 4096, L = 32), k/ε=256/0.102026.051.810.98998.896.510.5
Gated Subspace InferenceModel=OPT 6.7B (d = 4096, L = 32), k/ε=256/0.052026.051.820.99598.677.43.7
Baseline OPT 6.7BModel=OPT 6.7B (d = 4096, L = 32)2026.051.83————
FP16Model=LLaMA-2-7B, Setting=FP162026.065.471————
FP16Model=LLaMA-3.1-8B, Setting=FP162026.066.277————
SYM-INTModel=LLaMA-2-7B, Setting=W3A4 (GPTQ)2026.066.897————
Qift-MNZModel=LLaMA-2-7B, Setting=L=16 (GPTAQ)2026.067.122————
Qift-MNZModel=LLaMA-2-7B, Setting=Pure W2A4 (GPTAQ)2026.069.294————
SYM-INTModel=LLaMA-3.1-8B, Setting=W3A4 (GPTQ)2026.0610.954————
Qift-MNZModel=LLaMA-3.1-8B, Setting=L=16 (GPTAQ)2026.0611.936————
SYM-INTModel=LLaMA-2-7B, Setting=Pure W2A4 (GPTAQ)2026.0612.118————
Qift-MNZModel=LLaMA-3.1-8B, Setting=Pure W2A4 (GPTAQ)2026.0619.515————
SYM-INTModel=LLaMA-3.1-8B, Setting=Pure W2A4 (GPTAQ)2026.0629.695————