Bench Local Model Result
Model: gemma-4-E2B-it-Q8_0
Setting: Min_P = 1
Format & Constraints — Score: 95
- Format Constraints: 80
- Ordering and Sorting: 100
- Multi-Domain: 100
- Precision Under Pressure: 93
- Adversarial: 100
Information Extraction: 71
- Clean Extraction: 77
- Noisy and Informal: 69
- Multi-Entity: 80
- Implicit and Missing: 69
- Complex Documents: 59
Tool Selection — Score: 77
- Tool Selection: 83
- Parameter Precision: 100
- Multi-Step Chains: 67
- Restraint & Refusal: 83
- Error Recovery: 50
Reasoning & Logic — Score: 65
- Everyday Arithmetic: 77
- Logic Puzzles: 33
- Multi-Step Word Problems: 85
- Trick Questions and Traps: 77
- Applied Reasoning: 62
Conflicts & Overrides — Score: 60
- Hard System vs User (1/3): 33
- User Overrides Defaults (3/3): 100
- User vs Tool Data (3/3): 100
- Hard System vs Tool Data (1/3): 33
- Layered Conflicts (1/3): 33
Графики и визуализации
