List your company

Problem-solving on GSM8K (exact match %)

78.94Exact Match AccuracyDS2-INSTRUCT
9.946427.858245.7763.6818Sep 26, 2025Oct 24, 2025Nov 21, 2025Dec 19, 2025Jan 16, 2026Feb 13, 2026Mar 13, 2026

Evaluation results

MethodLinks
DS2-INSTRUCTModel Family=Qwen2.52026.0378.94
IA2 onlyBackbone=Qwen3-4B-Base, N=42025.0977.4
ExploreInstructModel Family=Qwen2.52026.0376.67
ICLBackbone=Qwen3-4B-Base, N=42025.0976.4
InstructMixModel Family=Qwen2.52026.0375.42
IA2 → SFTBackbone=Qwen3-4B-Base, N=42025.0973.6
Self-InstructModel Family=Qwen2.52026.0372.38
Zero-ShotModel Family=Qwen2.52026.0371.53
SFT onlyBackbone=Qwen3-4B-Base, N=42025.0970.9
DS2-INSTRUCTModel Family=Llama32026.0358.34
w/o ICLBackbone=Qwen3-4B-Base, N=42025.0956.4
DS2-INSTRUCTModel Family=Mistral2026.0342.75
InstructMixModel Family=Mistral2026.0338.26
InstructMixModel Family=Llama32026.0335.64
ExploreInstructModel Family=Llama32026.0333.81
Self-InstructModel Family=Llama32026.0319.83
Self-InstructModel Family=Mistral2026.0318.52
ExploreInstructModel Family=Mistral2026.0317.93
Zero-ShotModel Family=Mistral2026.0314.75
Zero-ShotModel Family=Llama32026.0312.6