List your company

Mathematical Reasoning on GSM8K, MATH, SVAMP, ASDiv, MAWPS, TAB, MQA, SAT (test)

56GSM8K AccuracyTAD
-0.1614.422943.58Feb 24, 2026

Evaluation results

MethodLinks
TADK=1, Teacher=Phi3-14B, Student=Llama2-7B, Data=OWM(2.5B) + ORCAMEL2026.025610.277.277.191.839.839.240.654
TADK=20, Teacher=Phi3-14B, Student=Llama2-7B, Data=OWM(2.5B) + ORCAMEL2026.0252.8877.676.992.439.23946.954.1
TADK=5, Teacher=Phi3-14B, Student=Llama2-7B, Data=OWM(2.5B) + ORCAMEL2026.0251.69.276.775.491.238.740.537.552.6
TADK=10, Teacher=Phi3-14B, Student=Llama2-7B, Data=OWM(2.5B) + ORCAMEL2026.0251.48.476.675.590.638.739.244.453.1
Vanilla KDTeacher=Phi3-14B, Student=Llama2-7B, Data=OWM(2.5B) + ORCAMEL2026.0250.58.175.374.490.529.737.234.450
WizardMath-7BData=+RL with Evol Instruct2026.0246.6756.865.281.13520.323.141.9
Orca2-7BData=+SFT (ORCA) + KTO2026.02406.270.26787.530.431.628.145.1
Llemma-7BData=+ProofPile(0.2T)2026.0239.715.456.967.783.34740.94449.4
TADK=1, Teacher=Phi3-4B, Student=TinyLlama-1.1B, Data=OWM(2.5B) + ORCAMEL2026.0236.86.867.867.981.725.416.328.141.4
Rho-1-Math(1.1B)Data=+OWM (30B)2026.0236.313.452.666.583.629.532.118.541.5
TADK=5, Teacher=Phi3-4B, Student=TinyLlama-1.1B, Data=OWM(2.5B) + ORCAMEL2026.0233.27.465.468.785.627.617.934.442.5
Vanilla KDTeacher=Phi3-4B, Student=TinyLlama-1.1B, Data=OWM(2.5B) + ORCAMEL2026.0230.86.864.662.580.720.116.727.538.7
TADK=10, Teacher=Phi3-4B, Student=TinyLlama-1.1B, Data=OWM(2.5B) + ORCAMEL2026.0230.1965.768.485.424.118.229.841.3
TADK=20, Teacher=Phi3-4B, Student=TinyLlama-1.1B, Data=OWM(2.5B) + ORCAMEL2026.0228.27.266.268.284.224.617.12540.1
Llama2-7BData=CLM + SFT + OWM(2.5B) + ORCAMEL2026.02224.247.756.372.337.72328.136.4
TinyLlama(TL)-1.1BData=CLM + SFT + OWM(2.5B) + ORCAMEL2026.0219.6449.458.874.321.81828.134.3
Llama2-7BData=Web (2T)2026.0214.23.639.151.663.630.912.532.831.4
TADTop-K (K)=5, Data (#Tkns)=+ OWM(2.5B), Teacher=Phi3 4B, Student=TinyLlama-1.1B, Distillation Strategy=Tail-Aware Distillation, Evaluation Protocol=few-shot CoT2026.027.14.819.235.946.715.91022.620.3
TADTop-K (K)=20, Data (#Tkns)=+ OWM(2.5B), Teacher=Phi3 4B, Student=TinyLlama-1.1B, Distillation Strategy=Tail-Aware Distillation, Evaluation Protocol=few-shot CoT2026.026.53.818.231.740.913.7931.219.4
Llama3.2-1.2B-PTData (#Tkns)=Web (9T), Evaluation Protocol=few-shot CoT2026.026.54.221.735.744.221.113.26.219.1
TADTop-K (K)=10, Data (#Tkns)=+ OWM(2.5B), Teacher=Phi3 4B, Student=TinyLlama-1.1B, Distillation Strategy=Tail-Aware Distillation, Evaluation Protocol=few-shot CoT2026.026.44.619.73342.712.99.337.520.7
Vanilla KDData (#Tkns)=+ OWM(2.5B), Teacher=Phi3 4B, Student=TinyLlama-1.1B, Distillation Strategy=Vanilla KD, Evaluation Protocol=few-shot CoT2026.026.14.221.133.541.515.511.216.718.7
TADTop-K (K)=1, Data (#Tkns)=+ OWM(2.5B), Teacher=Phi3 4B, Student=TinyLlama-1.1B, Distillation Strategy=Tail-Aware Distillation, Evaluation Protocol=few-shot CoT2026.026.16.222.133.141.51411.321.919.5
CLM (no KD)Data (#Tkns)=+ OWM(2.5B), Student=TinyLlama-1.1B, Evaluation Protocol=few-shot CoT2026.023.93.817.929.739.512.210.815.616.7
MiniPLMData (#Tkns)=+ OWM(2.5B), Evaluation Protocol=few-shot CoT2026.023.33.413.427.33410.810.512.514.4
Gemma3-1B-PTData (#Tkns)=Web (2T), Evaluation Protocol=few-shot CoT2026.022.12.212.817.122.411.114.515.612.2
TinyLlama(TL)-1.1BData=Web (2.5T)2026.0222.69.516.320.112.712.815.611.4
TinyLlama(TL)-1.1BData (#Tkns)=Web (2.5T), Evaluation Protocol=few-shot CoT2026.0222.69.516.320.112.712.815.611.4