List your company

Language Modeling on The Pile (test)

9.213PPL (The Pile Test)GPT-Neo 1.3B
0.9415256.77401112.6065168.43899May 19, 2023Nov 19, 2023May 22, 2024Nov 23, 2024May 26, 2025Nov 27, 2025May 31, 2026

Evaluation results

MethodLinks
GPT-Neo 1.3Btheta=0, Prefix Size=50, Suffix Size=50, Prompt Length=12023.059.213——————————————————————
BaseBudget=24x, Configuration=(24 ⊗ 1), Params=24x, FLOPs=24x2026.029.49——————————————————————
GPT-Neo 1.3Btheta=0.5, Prefix Size=50, Suffix Size=50, Prompt Length=12023.059.758——————————————————————
BaseBudget=12x, Configuration=(12 ⊗ 1), Params=12x, FLOPs=12x2026.029.98——————————————————————
GPT-Neo 1.3Btheta=0.75, Prefix Size=50, Suffix Size=50, Prompt Length=12023.0510.267——————————————————————
DAG-MoE-lNumber of Parameters=699M2026.0510.27——————————————————————
LoopFormerBudget=24x, Configuration=(3 ⊗ 8), Params=3x, FLOPs=24x2026.0210.28——————————————————————
TMLTBudget=24x, Configuration=(3 ⊗ 8), Params=3x, FLOPs=24x2026.0210.38——————————————————————
MoE-lNumber of Parameters=699M2026.0510.51——————————————————————
TMLT-EEBudget=24x, Configuration=(3 ⊗ 8), Params=3x, FLOPs=24x2026.0210.7——————————————————————
GPT-Neo 1.3Btheta=1, Prefix Size=50, Suffix Size=50, Prompt Length=12023.0510.775——————————————————————
Base-LoopBudget=24x, Configuration=(3 ⊗ 8), Params=3x, FLOPs=24x2026.0210.91——————————————————————
LoopFormerBudget=12x, Configuration=(3 ⊗ 4), Params=3x, FLOPs=12x2026.0211.12——————————————————————
BaseBudget=6x, Configuration=(6 ⊗ 1), Params=6x, FLOPs=6x2026.0211.13——————————————————————
Base-Loop-EE-ConsBudget=24x, Configuration=(3 ⊗ 8), Params=3x, FLOPs=24x2026.0211.56——————————————————————
Naive-Loop-EEBudget=24x, Configuration=(3 ⊗ 8), Params=3x, FLOPs=24x2026.0211.6——————————————————————
Naive-Loop-EEBudget=12x, Configuration=(3 ⊗ 4), Params=3x, FLOPs=12x2026.0211.66——————————————————————
Base-Loop-EE-ConsBudget=12x, Configuration=(3 ⊗ 4), Params=3x, FLOPs=12x2026.0212——————————————————————
TMLT-EEBudget=12x, Configuration=(3 ⊗ 4), Params=3x, FLOPs=12x2026.0212.18——————————————————————
Naive-Loop-EEBudget=6x, Configuration=(3 ⊗ 2), Params=3x, FLOPs=6x2026.0212.61——————————————————————
LoopFormerBudget=6x, Configuration=(3 ⊗ 2), Params=3x, FLOPs=6x2026.0214.3——————————————————————
Base-Loop-EE-ConsBudget=6x, Configuration=(3 ⊗ 2), Params=3x, FLOPs=6x2026.0215.07——————————————————————
GPT-Neo 125Mtheta=0, Prefix Size=50, Suffix Size=50, Prompt Length=12023.0515.71——————————————————————
TMLT-EEBudget=6x, Configuration=(3 ⊗ 2), Params=3x, FLOPs=6x2026.0215.79——————————————————————
GPT-Neo 125Mtheta=1.25, Prefix Size=50, Suffix Size=50, Prompt Length=12023.0516.601——————————————————————
GPT2 1.5BPrefix Size=50, Suffix Size=50, Prompt Length=12023.0517.155——————————————————————
GPT-Neo 125Mtheta=1.5, Prefix Size=50, Suffix Size=50, Prompt Length=12023.0517.499——————————————————————
GPT-Neo 125Mtheta=1.75, Prefix Size=50, Suffix Size=50, Prompt Length=12023.0519.691——————————————————————
SAGE(Hybrid)Model Scale=1.3B2026.0424.33———————————————————0.9——
LionModel Scale=0.6B2026.0426.58———————————————————2.2——
SAGE(Hybrid)Model Scale=0.6B2026.0426.71———————————————————0.7——
AdamWModel Scale=1.3B2026.0427.81———————————————————9.8——
LionModel Scale=1.3B2026.0428.37———————————————————4.9——
Lion-HybridModel Scale=1.3B2026.0428.4———————————————————0.9——
SinkGD(Hybrid)Model Scale=1.3B2026.0428.71———————————————————1.9——
Lion-HybridModel Scale=0.6B2026.0428.73———————————————————0.7——
SAGE(Hybrid)Model Scale=270M2026.0429.95———————————————————0.5——
LionModel Scale=270M2026.0430.24———————————————————1——
GPT2 124MPrefix Size=50, Suffix Size=50, Prompt Length=12023.0530.323——————————————————————
AdamWModel Scale=0.6B2026.0431.74———————————————————4.4——
Lion-HybridModel Scale=270M2026.0432.1———————————————————0.5——
SinkGD(Hybrid)Model Scale=0.6B2026.0432.85———————————————————1.5——
SinkGD(Hybrid)Model Scale=270M2026.0434.3———————————————————0.9——
AdamWModel Scale=270M2026.0437.35———————————————————2.1——
APOLLOModel Scale=1.3B2026.0439.54———————————————————3.6——
APOLLOModel Scale=0.6B2026.0444.84———————————————————2.1——
APOLLOModel Scale=270M2026.0452.83———————————————————1.2——
SinkGD-PureModel Scale=1.3B2026.0498.84———————————————————0——
SAGE-PureModel Scale=270M2026.04116———————————————————1——
SinkGD-PureModel Scale=0.6B2026.04117.8———————————————————0——
SAGE-PureModel Scale=0.6B2026.04179———————————————————2.2——
SinkGD-PureModel Scale=270M2026.04192.7———————————————————0——
SAGE-PureModel Scale=1.3B2026.04216———————————————————4.9——
GLM-130B2022.10———————————————————0.634———
Gopher2021.12—1.141.090.8990.6770.890.7120.7410.6910.6950.6560.6620.6410.590.5770.5460.5250.5130.377————
GPT-32022.10———————————————————0.742———
GPT-32021.12—1.370.9460.9320.6240.9750.8020.7790.6980.7231.160.8380.7730.6120.6250.5660.690.6120.645————
HullFTTime Budget (T)=1.75s, Average Selection Size (N*)=192026.05—————————————————————0.74483.83
HullFTTime Budget (T)=2.0s, Average Selection Size (N*)=222026.05—————————————————————0.73123.44
Jurassic2021.12—1.040.8570.879—0.8690.835—0.6690.7410.890.680.6550.590.5870.5370.5790.5140.358————
Jurassic-12022.10———————————————————0.65———
kNNTime Budget (T)=1.75s, Average Selection Size (N*)=152026.05—————————————————————0.7842—
kNNTime Budget (T)=2.0s, Average Selection Size (N*)=172026.05—————————————————————0.7763—
SIFTTime Budget (T)=1.75s, Average Selection Size (N*)=112026.05—————————————————————0.7831—
SIFTTime Budget (T)=2.0s, Average Selection Size (N*)=132026.05—————————————————————0.7656—