List your company
ResearchFormerly WizWand

Arithmetic Reasoning on Countdown (Accuracy)

63.1AccuracyGRPO + coupled rhythm credit
-2.347214.643931.63548.6261Sep 25, 2025Nov 3, 2025Dec 13, 2025Jan 22, 2026Mar 2, 2026Apr 11, 2026May 21, 2026
Updated 4mo ago

Evaluation results

MethodLinks
GRPO + coupled rhythm creditBackbone=Qwen3-4B-Base2025.1063.1
GRPO + global-anchor creditBackbone=Qwen3-4B-Base2025.1060.4
GRPO + local-chunk creditBackbone=Qwen3-4B-Base2025.1059.9
GRPO + high-entropy creditBackbone=Qwen3-4B-Base2025.1057.7
GRPO + random creditBackbone=Qwen3-4B-Base2025.1055
GRPOBackbone=Qwen3-4B-Base2025.1052.6
LIFT2Backbone=Dream-7B, H=22026.0533.6
Instruct Model + GIFTShots=5-shot, Training Dataset=s1K, Tuning Method=LoRA Tuning2025.0927.5
Instruct Model + GIFTShots=5-shot, Training Dataset=s1.1K, Tuning Method=LoRA Tuning2025.0926
LIFT3Backbone=Dream-7B, H=32026.0525.6
VanillaBackbone=Dream-7B2026.0525
GIFTBackbone=Dream-7B2026.0523.4
Instruct Model + SFTShots=5-shot, Training Dataset=s1K, Tuning Method=LoRA Tuning2025.0923.2
CARTBackbone=Dream-7B2026.0522.3
Instruct Model + SFTShots=5-shot, Training Dataset=s1.1K, Tuning Method=LoRA Tuning2025.0921.7
InstructBackbone=Dream-7B2026.0521.1
GIFTTraining Dataset=s1K, Tuning Method=LoRA Tuning, Base Model=LLaDA-8B-Instruct, Evaluation Protocol=0-shot2025.090.281
GIFTTraining Dataset=s1K-1.1, Tuning Method=LoRA Tuning, Base Model=LLaDA-8B-Instruct, Evaluation Protocol=0-shot2025.090.218
GIFTTraining Dataset=Tulu3-10k, Tuning Method=Full Parameter Tuning, Base Model=LLaDA-8B-Instruct, Evaluation Protocol=0-shot2025.090.213
SFTTraining Dataset=s1K, Tuning Method=LoRA Tuning, Base Model=LLaDA-8B-Instruct, Evaluation Protocol=0-shot2025.090.211
SFTTraining Dataset=s1K-1.1, Tuning Method=LoRA Tuning, Base Model=LLaDA-8B-Instruct, Evaluation Protocol=0-shot2025.090.207
GIFTTraining Dataset=openr1-3k, Tuning Method=LoRA Tuning, Base Model=LLaDA-8B-Instruct, Evaluation Protocol=0-shot2025.090.188
SFTTraining Dataset=Tulu3-10k, Tuning Method=Full Parameter Tuning, Base Model=LLaDA-8B-Instruct, Evaluation Protocol=0-shot2025.090.182
SFTTraining Dataset=openr1-3k, Tuning Method=LoRA Tuning, Base Model=LLaDA-8B-Instruct, Evaluation Protocol=0-shot2025.090.173
LLaDA-8B-InstructEvaluation Protocol=0-shot2025.090.17