Phi 4 Mini Instruct is a lightweight (3.8B parameters) open model built upon synthetic data and filtered web data, focusing on high-quality reasoning. It supports a 128K token context length and is enhanced for instruction adherence and safety via supervised fine-tuning and direct preference optimization.
发布日期2025年2月1日
参数规模3.8B
上下文长度128K
许可证MIT
知识截止2024年6月1日
Benchmarks
评测成绩
| 评测基准 | 类别 | 分数 | 来源 |
|---|---|---|---|
| GSM8k | math reasoning | 88.6 | 来源 |
| ARC-C | reasoning general | 83.7 | 来源 |
| BoolQ | language reasoning | 81.2 | 来源 |
| OpenBookQA | reasoning general | 79.2 | 来源 |
| PIQA | reasoning physics general | 77.6 | 来源 |
| Social IQa | reasoning psychology | 72.5 | 来源 |
| BIG-Bench Hard | reasoning math language | 70.4 | 来源 |
| HellaSwag | reasoning | 69.1 | 来源 |
| MMLU | general reasoning language math | 67.3 | 来源 |
| Winogrande | reasoning language | 67.0 | 来源 |
| TruthfulQA | general reasoning legal healthcare finance | 66.4 | 来源 |
| MATH | math reasoning | 64.0 | 来源 |
| MGSM | math reasoning | 63.9 | 来源 |
| MMLU-Pro | language reasoning math general | 52.8 | 来源 |
| Multilingual MMLU | general reasoning language | 49.3 | 来源 |
| Arena Hard | general reasoning creativity | 32.8 | 来源 |
| GPQA | reasoning general | 25.2 | 来源 |
Pricing
API 价格对比
| 服务商 | 输入价 | 输出价 | 上下文 | 吞吐(tok/s) | 延迟(s) | 函数调用 | 代码执行 | 联网搜索 |
|---|---|---|---|---|---|---|---|---|
| Azure | $0.08 | $0.30 | 128K | — | — | ✓ | ✗ | ✗ |
| Azure Cognitive Services | $0.08 | $0.30 | 128K | — | — | ✓ | ✗ | ✗ |
价格单位：美元/百万 token，数据来自社区整理，仅供参考。