List your company

Long-context language modeling on InfiniteBench (Accuracy and Sparsity)

46.19Code Debug AccuracyFast-dLLM v2
-1.847610.623723.09535.5663Feb 26, 2026Mar 15, 2026Apr 1, 2026Apr 18, 2026May 5, 2026May 22, 2026Jun 9, 2026

Evaluation results

MethodLinks
Fast-dLLM v2Base Model=Dream-v0-Base-7B2026.0646.19———————————24.29—27.12—27.12—30.32—057.21
S2OModel backbone=Llama-3.1-8B2026.0224.41791.7816.692.227.192.251291.51191.0690.9634.391.5498.693.999.893.8237.992.11——
FullModel backbone=Llama-3.1-8B2026.0224.116.7018.7027.1013.5013.20033.1099.3099.3038.30——
XAttnModel backbone=Llama-3.1-8B2026.0223.41871.0616.571.3222.371.2813.572.47868.3173.0332.970.7599.868.9299.568.5637.170.63——
PBSModel backbone=Llama-3.1-8B2026.0222.817.975.3919.275.7824.575.711.577.0912.772.9277.6632.975.1383.174.1599.273.763675.29——
FlexPrefillModel backbone=Llama-3.1-8B2026.0221.816.785.9117.885.2324.585.3412.584.1611.384.883.7632.886.8210087.698.885.137.485.41——
XAttnModel backbone=Qwen3-8B2026.022016.466.311.967.0141.566.9716.569.143.766.6971.4730.661.8910062.298162.1234.665.99——
S2OModel backbone=Qwen3-8B2026.022017.590.85290.8944.589.82886.693.886.6387.4532.387.3610085.5499.790.3338.688.39——
FullModel backbone=Qwen3-8B2026.0218.51802.2048.9027.503.80031.70100096.9038.60——
PREFILLING-DLLMBase Model=Dream-v0-Base-7B, Sparsity=inter + intra-sparsity, B=5122026.0618.02———————————24.57—95.42—70—43.62—053.71
PBSModel backbone=Qwen3-8B2026.0217.21877.31.974.6847.674.7330.576.493.874.2478.0633.770.9810071.7510072.0939.275.04——
FlexPrefillModel backbone=Qwen3-8B2026.021714.982.51.982.664584.941683.293.983.2885.813481.6499.881.819982.3736.883.14——
Fast-dLLMBase Model=Dream-v0-Base-7B2026.068.38———————————3.43—0.17—3.05—3.96—08.73
YaRNBase Model=Dream-v0-Base-7B, Context Window=128K2026.067.87———————————3.43—0—0—2.9—06.11
Sparse-dLLMBase Model=Dream-v0-Base-7B, B=5122026.063.81———————————3.43—39.32—25.59—12.24—01.31
dKV-CacheBase Model=Dream-v0-Base-7B, B=5122026.063.05———————————4—0.51—0—2.21—05.68
Sparse-dLLMBase Model=UltraLLaDA, B=5122026.061.52———————————7.43—8.14—2.37—4.12—05.24
PREFILLING-DLLMBase Model=UltraLLaDA, Sparsity=inter + intra-sparsity, B=5122026.061.52———————————26—13.22—30—14.19—014.41
VanillaBase Model=UltraLLaDA, Context Window=128K2026.060.85———————————19.15—10.81—1.72—7.1—010.09
Fast-dLLMBase Model=UltraLLaDA2026.060.25———————————14.29—16.1—6.61—7.52—07.86
dKV-CacheBase Model=UltraLLaDA, B=5122026.060.25———————————16—5.59—7.12—6.21—08.3
VanillaBase Model=Dream-v0-Base-7B, Context Window=2K2026.060———————————34.57—1.69—2.54—12.44—035.81