Language Understanding and Knowledge on MMLU & C-EVAL Suite
72.2MMLUQwen2.5-7B + Inf-1M
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen2.5-7B + Inf-1MBase Model Architecture=Qwen2.5-7B, Fine-tuning Dataset=Inf-1M2025.12 | 72.2 | 79.4 | 75.8 | — | |
| DataFlow-SFT-15K(filtered)Data Source=DataFlow-SFT-15K, Filtering Method=DataFlow SFT pipeline, SFT Data Scale=15K2025.12 | 72.2 | 80.4 | — | 76.3 | |
| Qwen2.5-7B + DataFlow-Instruct-10KBase Model Architecture=Qwen2.5-7B, Fine-tuning Dataset=DataFlow-Instruct-10K2025.12 | 72.1 | 80.2 | 76.2 | — | |
| DataFlow-SFT-15K(random)Data Source=DataFlow-SFT-15K, Filtering Method=random, SFT Data Scale=15K2025.12 | 72.1 | 80 | — | 76.1 | |
| Qwen2.5-7B-BaseBase Model Architecture=Qwen2.5-7B, Fine-tuning Dataset=None (Base)2025.12 | 71.9 | 80 | 76 | — | |
| WizardLM(filtered)Data Source=WizardLM, Filtering Method=DataFlow SFT pipeline, SFT Data Scale=5K2025.12 | 71.9 | 79.6 | — | 75.8 | |
| Qwen2.5-7B + Inf-10KBase Model Architecture=Qwen2.5-7B, Fine-tuning Dataset=Inf-10K2025.12 | 71.8 | 79.9 | 75.8 | — | |
| Qwen2.5-7B-InstructBase Model Architecture=Qwen2.5-7B, Fine-tuning Dataset=Instruct2025.12 | 71.8 | 79.6 | 75.7 | — | |
| Alpaca(random)Data Source=Alpaca, Filtering Method=random, SFT Data Scale=5K2025.12 | 71.8 | 80 | — | 75.9 | |
| Alpaca(filtered)Data Source=Alpaca, Filtering Method=DataFlow SFT pipeline, SFT Data Scale=5K2025.12 | 71.8 | 80 | — | 75.9 | |
| WizardLM(random)Data Source=WizardLM, Filtering Method=random, SFT Data Scale=5K2025.12 | 71.8 | 79.2 | — | 75.5 | |
| Qwen2-7B-InstructBase Model Architecture=Qwen2-7B, Fine-tuning Dataset=Instruct2025.12 | 69.9 | 82 | 76 | — | |
| Qwen2-7B-BaseBase Model Architecture=Qwen2-7B, Fine-tuning Dataset=None (Base)2025.12 | 69.6 | 82.8 | 76.2 | — | |
| Qwen2-7B + Inf-1MBase Model Architecture=Qwen2-7B, Fine-tuning Dataset=Inf-1M2025.12 | 69.5 | 83 | 76.2 | — | |
| Qwen2-7B + DataFlow-Instruct-10KBase Model Architecture=Qwen2-7B, Fine-tuning Dataset=DataFlow-Instruct-10K2025.12 | 69.4 | 82.8 | 76.1 | — | |
| Qwen2-7B + Inf-10KBase Model Architecture=Qwen2-7B, Fine-tuning Dataset=Inf-10K2025.12 | 69.3 | 83 | 76.2 | — |