List your company

Mathematical Reasoning on GSM8K (Accuracy, Time, and Speed Up)

13.4Speed Up (x)ES-dLLM
2.3765.2388.110.962May 30, 2024Sep 15, 2024Jan 1, 2025Apr 19, 2025Aug 5, 2025Nov 21, 2025Mar 10, 2026

Evaluation results

MethodLinks
ES-dLLMBase Model=Dream-7B-Base, Few-shot=52026.0313.4———————282.67——————73.92
Fast-DualModel=Dream 7B, nu=12026.0312.19—68.39——————153.21——————
GKTTeacher=Llama2-70B, Student=Llama2-7B, Output Length=40-3002024.0510.727.8128.5814.1813,440.71———————————
DualCacheBase Model=Dream-7B-Base, Few-shot=52026.0310.2———————211.67——————73.16
Llama2-13BModel Type=Single Model, Output Length=3002024.0510.13—23.659.2514,215.12———————————
DyLLMModel=Dream 7B, Threshold (τ)=0.995, nu=12026.039.6—78.09——————120.62——————
DyLLMModel=Dream 7B, Threshold (τ)=0.9975, nu=12026.038.89—79.3——————111.79——————
GKTTeacher=Llama2-70B, Student=Llama2-13B, Output Length=40-3002024.058.867.8135.4111.7616,250.34———————————
DyLLMModel=LLaDA 8B, Threshold (τ)=0.99, nu=12026.037.6—79.08——————87.21——————
DyLLMModel=LLaDA 8B, Threshold (τ)=0.995, nu=12026.036.99—78.01——————80.15——————
Fast-DualModel=LLaDA 8B, nu=12026.036.56—78.24——————75.24——————
DraftDraft-Target Model Pair=Llama 3.1 70B – Llama 3.2 1B2025.126.08————————99.919—0.375————
iCoTBase Model=Qwen2.5-Base-1.5B2026.025.71—34—————————————
Window-DiffusionModel Variant=Base, Window Size (L)=16, Refresh Cycle=32, Early Stopping=Disabled2026.015.7—76.1——————29.9——————
No CoTBase Model=Qwen2.5-Base-1.5B2026.025.59—34.11—————————————
JotBackbone=Dream-7B, tau_max=90, tau_min=1, gamma=0.5, D=82026.025.54——————————78.8————
EAGLE3 + DMModel=Llama-3.3-70B-Instruct, Draft length (L)=92026.025.03—80———————7.48—————
Eagle3+LTDModel=DeepSeek-R1-Distill-LLaMA 8B, Temperature=0, Decoding Mode=Greedy2026.034.98—————————8.53—————
EAGLE3 + DMModel=Llama-3.3-70B-Instruct, Draft length (L)=72026.024.86—81.25———————7.07—————
EAGLE3Model=Llama-3.3-70B-Instruct, Draft length (L)=92026.024.68—78.75———————6.71—————
EAGLE3Model=Llama-3.3-70B-Instruct, Draft length (L)=72026.024.65—78.75———————6.71—————
Eagle3+LTDModel=Vicuna-13B-v1.3, Temperature=0, Decoding Mode=Greedy2026.034.59—————————7.38—————
DAWNModel=Dream-v0-Instruct-7B2026.024.52—73.16—————32.99———————
LocalLeapModel=Dream-v0-Instruct-7B2026.024.51—73.24—————32.94———————
Fast-dLLMModel Variant=Base, Caching Strategy=Dual-Cache2026.014.5—74.4——————24.1——————
DAWNModel=LLaDA-1.52026.024.48—80.82—————43.24———————
Fast-PrefixModel=LLaDA 8B, nu=12026.034.45—77.18——————51.05——————
TALONModel=DSL-8B, Temperature=02026.014.43——————————7.46————
LocalLeapModel=LLaDA-1.52026.024.41—80.14—————42.6———————
DOUBLEBase Model=LLaMA3.3-70B2026.014.41————————————10.86——
DAWNModel=LLaDA-8B-Instruct2026.024.33—77.94—————44.72———————
EAGLE3 + DMModel=Llama-3.3-70B-Instruct, Draft length (L)=52026.024.32—78.75———————6.09—————
LocalLeapModel=LLaDA-8B-Instruct2026.024.28—77.33—————44.19———————
EAGLE-3Model=DSL-8B, Temperature=02026.014.24——————————7.4————
EAGLE3Base Model=LLaMA3.3-70B2026.014.21————————————6.02——
JotModel=Dream-7B2026.024.21———————————78.8———
Fast-dLLMModel Variant=Base, Caching Strategy=Prefix-Cache2026.014.2—74.7——————22.5——————
EAGLE3Model=Llama-3.3-70B-Instruct, Draft length (L)=52026.024.19—78.75———————5.82—————
Eagle3+GSModel=DeepSeek-R1-Distill-LLaMA 8B, Temperature=12026.034.14—————————7.64—————
Eagle3+LTDModel=DeepSeek-R1-Distill-LLaMA 8B, Temperature=12026.034.02—————————7.95—————
TALONModel=Vicuna-13B, Temperature=02026.013.99——————————6.8————
Eagle3Model=DeepSeek-R1-Distill-LLaMA 8B, Temperature=12026.033.97—————————6.89—————
ConfidenceModel=Dream-v0-Instruct-7B2026.023.95—75.2—————28.81———————
EAGLE3Base Model=LLaMA3.3-70B, Temperature=1, Decoding Setting=non-greedy2026.013.91———————————5.6———
DOUBLEBase Model=LLaMA3.3-70B, Temperature=1, Decoding Setting=non-greedy2026.013.89———————————9.58———
EAGLE-3Model=Vicuna-13B, Temperature=02026.013.87——————————6.74————
DOUBLEBase Model=Deepseek-33B2026.013.83————————————10.86——
GKTTeacher=Llama2-13B, Student=Bloom-7B, Output Length=40-3002024.053.824.6210.315.913,275.9———————————
TALONModel=Llama3-8B, Temperature=02026.013.81——————————6.24————
Eagle3+GSModel=Vicuna-13B-v1.3, Temperature=12026.033.81—————————6.39—————
Eagle3+LTDModel=Llama-3.1-8B-Instruct, Temperature=0, Decoding Mode=Greedy2026.033.79—————————6.73—————
JotBackbone=LLaDA-8B, tau_max=30, tau_min=1, gamma=0.5, D=82026.023.75——————————73.4————
Eagle3+LTDModel=Vicuna-13B-v1.3, Temperature=12026.033.75—————————6.38—————
FailFastTarget Model=Qwen2.5-32B-Instruct2025.123.71———————————————
Eagle3Model=Vicuna-13B-v1.3, Temperature=12026.033.68—————————5.64—————
Eagle3+DiscoModel=Vicuna-13B-v1.3, Temperature=12026.033.67—————————5.74—————
dLLM-CacheModel=Dream 7B, nu=12026.033.67—72.4——————46.19——————
OPT-TreeModel=Vicuna-13B, Temperature=02026.013.65——————————6.77————
Eagle3+Spec++Model=Vicuna-13B-v1.3, Temperature=12026.033.64—————————5.63—————
EAGLE-3Model=Llama3-8B, Temperature=02026.013.63——————————6.42————
Eagle3+DDDModel=Vicuna-13B-v1.3, Temperature=12026.033.58—————————5.46—————
PEARLBase Model=LLaMA3.3-70B2026.013.52————————————8.57——
Fast-PrefixModel=Dream 7B, nu=12026.033.45—74.45——————43.32——————
Eagle3Model=Llama-3.1-8B-Instruct, Temperature=12026.033.43—————————5.24—————
STDDBackbone=Dream-7B2025.123.41—75——————30.3——————
Eagle3+C2TModel=Vicuna-13B-v1.3, Temperature=12026.033.41—————————5.63—————
GKTTeacher=Llama2-13B, Student=Bloom-3B, Output Length=40-3002024.053.44.627.735.384,186.22———————————
ConfidenceModel=LLaDA-1.52026.023.37—80.74—————32.49———————
Eagle3+GSModel=Llama-3.1-8B-Instruct, Temperature=12026.033.36—————————5.5—————
Eagle3+DDDModel=Llama-3.1-8B-Instruct, Temperature=12026.033.35—————————4.09—————
Eagle3+LTDModel=Llama-3.1-8B-Instruct, Temperature=12026.033.35—————————5.09—————
Eagle3+Spec++Model=Llama-3.1-8B-Instruct, Temperature=12026.033.33—————————4.74—————
Eagle3+GTModel=Vicuna-13B-v1.3, Temperature=12026.033.32—————————5.26—————
PEARLBase Model=LLaMA3.3-70B, Temperature=1, Decoding Setting=non-greedy2026.013.31———————————8.06———
Eagle3+GTModel=Llama-3.1-8B-Instruct, Temperature=12026.033.28—————————4.27—————
Eagle3+C2TModel=Llama-3.1-8B-Instruct, Temperature=12026.033.28—————————5.24—————
Eagle3+DiscoModel=Llama-3.1-8B-Instruct, Temperature=12026.033.26—————————5.08—————
Eagle3+SvipModel=Vicuna-13B-v1.3, Temperature=12026.033.26—————————5.78—————
ConfidenceModel=LLaDA-8B-Instruct2026.023.24—78.39—————33.44———————
Eagle3+SvipModel=Llama-3.1-8B-Instruct, Temperature=12026.033.21—————————4.92—————
dLLM-CacheModel=LLaDA 8B, nu=12026.033.21—77.18——————36.77——————
CoconutBase Model=Qwen2.5-Base-1.5B2026.023.14—32.65—————————————
Fast-dLLMTarget Model=Qwen2.5-32B-Instruct2025.123.1———————————————
PACER + OuroborosModels=Deepseek-Coder 1.3B/33B2026.023.09————————56.315.99—————
STDDBackbone=LLaDA-8B2025.123.07—83.1——————16.32——————
JotModel=LLaDA-8B2026.023.07———————————73.4———
EDA (Ours)Temperature=0, Target Model=Qwen2.5-Math-7B2026.033.06—————————4.79—————
FailFastTarget Model=Qwen2.5-14B-Instruct2025.123.04———————————————
Eagle3+LTDModel=Qwen3-32B, Temperature=0, Decoding Mode=Greedy2026.033.02—————————3.9—————
OuroborosModels=Deepseek-Coder 1.3B/33B2026.022.89————————52.664.43—————
EDA (Base)Temperature=0, Target Model=Qwen2.5-Math-7B2026.032.89—————————4.4—————
OuroborosBase Model=Deepseek-33B2026.012.88————————————5.95——
Full-FTTemperature=0, Target Model=Qwen2.5-Math-7B2026.032.88—————————4.37—————
LoRATemperature=0, Target Model=Qwen2.5-Math-7B2026.032.84—————————4.32—————
DOUBLEBase Model=LLaMA2-70B2026.012.83————————————7.86——
Eagle3+LTDModel=Qwen3 32B, Temperature=12026.032.83—————————3.76—————
PEARLBase Model=Deepseek-33B2026.012.82————————————7.83——
AR Draft ModelTarget Model=Qwen2.5-32B-Instruct2025.122.82———————————————
Eagle3+LTDModel=Qwen3-14B, Temperature=0, Decoding Mode=Greedy2026.032.82—————————3.92—————
DKV-CacheModel Variant=Base2026.012.8—74.1——————14.8——————