List your company

Inference Performance and Accuracy for Long-Context Summarization on GovReport

3.09SpeedupGraft
0.58361.23431.8852.5357Feb 23, 2026Mar 9, 2026Mar 23, 2026Apr 6, 2026Apr 20, 2026May 4, 2026May 19, 2026

Evaluation results

MethodLinks
GraftBackbone=LLaMA3.1-8B2026.053.09——5.95
EAGLE3-64KBackbone=LLaMA3.1-8B2026.052.84——5.81
GraftBackbone=Qwen3-14B2026.052.09——3.46
EAGLE3-64KBackbone=Qwen3-14B2026.051.83——3.36
GraftBackbone=Qwen3-8B2026.051.77——3.87
GraftBackbone=Qwen3-4B2026.051.68——3.23
TriforceBackbone=LLaMA3.1-8B2026.051.64——2.41
EAGLE3-64KBackbone=Qwen3-8B2026.051.59——3.79
TokenSwiftBackbone=LLaMA3.1-8B2026.051.58——1.74
KnapSpecBackbone=Llama3.1-70B, Layers (L)=802026.021.471394.4—
EAGLE3-64KBackbone=Qwen3-4B2026.051.44——3.12
SWIFTBackbone=Llama3.1-70B, Layers (L)=802026.021.338.6291.6—
KnapSpecBackbone=Llama3.1-8B, Layers (L)=322026.021.282597—
KnapSpecBackbone=Llama3.2-3B, Layers (L)=282026.021.2436.4795.2—
CLaSpBackbone=Llama3.1-70B, Layers (L)=802026.021.226.8293.4—
CLaSpBackbone=Llama3.1-8B, Layers (L)=322026.021.1114.1591.7—
KnapSpecBackbone=Llama3.2-1B, Layers (L)=162026.021.1158.7996.1—
SWIFTBackbone=Llama3.2-3B, Layers (L)=282026.021.0818.4564—
CLaSpBackbone=Llama3.2-3B, Layers (L)=282026.021.0818.9185.7—
SWIFTBackbone=Llama3.1-8B, Layers (L)=322026.021.0513.7262.1—
SWIFTBackbone=Llama3.2-1B, Layers (L)=162026.021.0331.3283.8—
CLaSpBackbone=Llama3.2-1B, Layers (L)=162026.021.0233.7583.6—
ARBackbone=Llama3.1-70B, Layers (L)=802026.0216.75——
ARBackbone=Llama3.1-8B, Layers (L)=322026.02115.71——
ARBackbone=Llama3.2-3B, Layers (L)=282026.02122.58——
ARBackbone=Llama3.2-1B, Layers (L)=162026.02146.56——
DELBackbone=Llama3.2-1B, Layers (L)=162026.020.8949.3155.6—
DELBackbone=Llama3.1-70B, Layers (L)=802026.020.876.620.1—
DELBackbone=Llama3.2-3B, Layers (L)=282026.020.8423.7236—
DELBackbone=Llama3.1-8B, Layers (L)=322026.020.7415.720.1—
MagicDecBackbone=LLaMA3.1-8B2026.050.68——2.05