Qwen2.5-32B-Instruct is an instruction-tuned 32 billion parameter language model, part of the Qwen2.5 series. It is designed to follow instructions, generate long texts (over 8K tokens), understand structured data (e.g., tables), and generate structured outputs, especially JSON. The model supports multilingual capabilities across over 29 languages.
发布日期2024年9月19日
参数规模32.5B
上下文长度—
许可证Apache 2.0
知识截止—
Benchmarks
评测成绩
| 评测基准 | 类别 | 分数 | 来源 |
|---|---|---|---|
| GSM8k | math reasoning | 95.9 | 来源 |
| 安全防御榜·Toxicity | security | 93.9 | — |
| 安全防御榜·Bias | security | 91.4 | — |
| 安全防御榜·Info Safety | security | 89.9 | — |
| 安全防御榜·Malicious Use | security | 88.6 | — |
| HumanEval | reasoning code | 88.4 | 来源 |
| 安全防御榜 | security | 88.3 | — |
| 安全防御榜·Human Rights | security | 87.5 | — |
| HellaSwag | reasoning | 85.2 | 来源 |
| BBH | reasoning math language | 84.5 | 来源 |
| 安全防御榜·Socioeconomic | security | 84.4 | — |
| MBPP | reasoning general | 84.0 | 来源 |
| MMLU-Redux | language reasoning math general | 83.9 | 来源 |
| MMLU | general reasoning language math | 83.3 | 来源 |
| MATH | math reasoning | 83.1 | 来源 |
| 安全防御榜·Child Safety | security | 82.4 | — |
| Winogrande | reasoning language | 82.0 | 来源 |
| 价值对齐榜·Fairness | security | 81.3 | — |
| MMLU-STEM | math reasoning physics chemistry | 80.9 | 来源 |
| 价值对齐榜·Morality | security | 79.0 | — |
| MultiPL-E | general language | 75.4 | 来源 |
| 价值对齐榜·Legality | security | 73.3 | — |
| 价值对齐榜·Safety | security | 72.7 | — |
| 价值对齐榜 | security | 72.2 | — |
| ARC-C | reasoning general | 70.4 | 来源 |
| MMLU-Pro | language reasoning math general | 69.0 | 来源 |
| MBPP+ | reasoning general | 67.2 | 来源 |
| TruthfulQA | general reasoning legal healthcare finance | 57.8 | 来源 |
| 价值对齐榜·Data Protection | security | 54.6 | — |
| HumanEval+ | reasoning | 52.4 | 来源 |
| GPQA | reasoning general | 49.5 | 来源 |
| TheoremQA | math reasoning physics finance | 44.1 | 来源 |
Pricing
API 价格对比
暂无 API 价格。