List your company

Mathematical Reasoning on GSM8K (Accuracy, Tokens, ActRatio)

103.69TokensTruncation
71.1976290.5213509.845729.1687Aug 7, 2025Sep 7, 2025Oct 8, 2025Nov 8, 2025Dec 9, 2025Jan 9, 2026Feb 9, 2026

Evaluation results

MethodLinks
TruncationBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.52025.08103.690.497
ASCoTBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.52025.08118.570.5579.5
ASCoTBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.62025.08128.380.681.6
TruncationBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.72025.08149.970.725.9
ASCoTBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.72025.08151.320.7183.7
ASCoTBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.82025.08169.960.7985.1
PromptBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.52025.08188.820.8983.7
ASCoTBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.92025.08197.650.9286.1
TruncationBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.92025.08202.060.9570.2
PromptBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.72025.08209.390.9884.9
Extra-CoT (CHRPO)Ratio=<POLICY>, Backbone=Qwen3-1.7B2026.022100.2485.8
OriginalBackbone=LLaMA-3.1-8B-Instruct, Ratio=1.02025.08213.17186.2
ASCoTBackbone=LLaMA-3.1-8B-Instruct, Ratio=1.02025.08214.24186.9
PromptBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.92025.08226.371.0684.1
TokenSkipRatio=0.2, Backbone=Qwen3-1.7B2026.022730.359.1
Extra-CoTRatio=0.2, Backbone=Qwen3-1.7B2026.023030.3480.2
Thinkless* (DeGRPO)Ratio=–, Backbone=Qwen3-1.7B2026.023560.4185.5
Extra-CoTRatio=0.4, Backbone=Qwen3-1.7B2026.024690.5382.3
TokenSkipRatio=<POLICY>, Backbone=Qwen3-1.7B2026.024730.5270
TokenSkipRatio=0.4, Backbone=Qwen3-1.7B2026.025160.5674.2
Extra-CoTRatio=<POLICY>, Backbone=Qwen3-1.7B2026.025690.6586.3
Extra-CoTRatio=0.6, Backbone=Qwen3-1.7B2026.026410.7385.5
TruncationRatio=0.6, Backbone=Qwen3-1.7B2026.026920.881.7
LC-PromptRatio=0.6, Backbone=Qwen3-1.7B2026.027630.8886.2
TokenSkipRatio=0.6, Backbone=Qwen3-1.7B2026.027700.8478.5
TokenSkipRatio=0.8, Backbone=Qwen3-1.7B2026.027940.9184.3
Extra-CoTRatio=0.8, Backbone=Qwen3-1.7B2026.028070.9286.7
Base ModelBackbone=Qwen3-1.7B2026.02873—86.8
Extra-CoTRatio=1.0, Backbone=Qwen3-1.7B2026.029021.0386.7
TokenSkipRatio=1.0, Backbone=Qwen3-1.7B2026.029161.0583.7