Mathematical Reasoning on Orca-Math (test)
35.2AccuracyQueryable LoRA
Evaluation Results
| Method | Links | |
|---|---|---|
| Queryable LoRABackbone=Mistral7B, Evaluation Protocol=Test Accuracy2026.05 | 35.2 | |
| Instruction-Queryable LoRAModel=Qwen/Qwen3-0.6B2026.05 | 34.4 | |
| Queryable LoRABackbone=Qwen0.5B, Evaluation Protocol=Test Accuracy2026.05 | 34.4 | |
| Instruction-Queryable LoRABackbone=Qwen0.5B, Evaluation Protocol=Test Accuracy2026.05 | 34.4 | |
| Instruction-Queryable LoRABackbone=Mistral7B, Evaluation Protocol=Test Accuracy2026.05 | 32.8 | |
| LoRABackbone=Mistral7B, Evaluation Protocol=Test Accuracy2026.05 | 32 | |
| SFTSetting=SFT (3,000 labels), Base Model=Qwen2.5-0.5B-Instruct, Labeled=3,000, Unlabeled=0, Filtered=0, Total=3,0002026.06 | 30.6 | |
| LoRABackbone=Qwen0.5B, Evaluation Protocol=Test Accuracy2026.05 | 30.4 | |
| RepLoRABackbone=Qwen0.5B, Evaluation Protocol=Test Accuracy2026.05 | 30.4 | |
| Proposed Semi-Supervised FrameworkSetting=Proposed (200 labels + 100k unlabeled), Base Model=Qwen2.5-0.5B-Instruct, Labeled=200, Unlabeled=100,000, Filtered=2,911, Total=3,1112026.06 | 30.1 | |
| RepLoRABackbone=Mistral7B, Evaluation Protocol=Test Accuracy2026.05 | 28.9 | |
| Queryable LoRAModel=amd/ReasonLite-0.6B-Turbo2026.05 | 28.1 | |
| Instruction-Queryable LoRAModel=amd/ReasonLite-0.6B-Turbo2026.05 | 28.1 | |
| Instruction-Queryable LoRAModel=LiquidAI/LFM2-700M2026.05 | 26.6 | |
| SFTSetting=SFT (200 labels), Base Model=Qwen2.5-0.5B-Instruct, Labeled=200, Unlabeled=0, Filtered=0, Total=2002026.06 | 26.6 | |
| Zero-shotSetting=Zero-shot, Base Model=Qwen2.5-0.5B-Instruct, Labeled=–, Unlabeled=-, Filtered=-, Total=-2026.06 | 25.2 | |
| LoRAModel=LiquidAI/LFM2-700M2026.05 | 25 | |
| LoRAModel=Qwen/Qwen3-0.6B2026.05 | 25 | |
| LoRAModel=amd/ReasonLite-0.6B-Turbo2026.05 | 23.4 | |
| DoRABackbone=Qwen0.5B, Evaluation Protocol=Test Accuracy2026.05 | 23.4 | |
| Queryable LoRAModel=LiquidAI/LFM2-700M2026.05 | 21.9 | |
| Instruction-Queryable LoRAModel=amd/ReasonLite-0.6B2026.05 | 21.9 | |
| Queryable LoRAModel=Qwen/Qwen3-0.6B2026.05 | 20.3 | |
| LoRAModel=amd/ReasonLite-0.6B2026.05 | 20.3 | |
| Queryable LoRAModel=amd/ReasonLite-0.6B2026.05 | 20.3 | |
| DoRABackbone=Mistral7B, Evaluation Protocol=Test Accuracy2026.05 | 20.3 | |
| Queryable LoRAModel=ibm-granite/granite-4.0-350m2026.05 | 17.2 | |
| Queryable LoRAModel=Qwen/Qwen2.5-Coder-0.5B-Instruct2026.05 | 14.1 | |
| Instruction-Queryable LoRAModel=ibm-granite/granite-4.0-350m2026.05 | 10.9 | |
| Queryable LoRAModel=LiquidAI/LFM2.5-350M2026.05 | 9.4 | |
| LoRAModel=Qwen/Qwen2.5-Coder-0.5B-Instruct2026.05 | 9.4 | |
| Instruction-Queryable LoRAModel=Qwen/Qwen2.5-Coder-0.5B-Instruct2026.05 | 9.4 | |
| LoRAModel=ibm-granite/granite-4.0-350m2026.05 | 7.8 | |
| LoRAModel=LiquidAI/LFM2.5-350M2026.05 | 6.2 | |
| Instruction-Queryable LoRAModel=LiquidAI/LFM2.5-350M2026.05 | 6.2 |