Microsoft
Phi-4 14B
当前
874.9K下载量2.2K点赞Dec 2024发布日期16K tokens上下文MIT许可证75 优秀质量
Phi-4 14B (14B parameters) requires approximately 13.4 GB of VRAM with Q4_K_M quantization. For the best balance of quality and speed, we recommend hardware with at least 16 GB of VRAM.
快速开始
— 复制粘贴即可本地运行Copy-paste commands to run Phi-4 14B on your machine.
Run
ollama run phi4Quick specs
Parameters14B
Architecturedense
Context16K tokens
Modalitytext
Min RAM5.5 GB
Rec. RAM8.5 GB (Q4_K_M)
LicenseMIT
FamilyPhi
✓ Chat✓ Reasoning
About this model
- •Publicly available documents filtered rigorously for quality, selected high-quality educational data, and code
- •Newly created synthetic, “textbook-like” data for the purpose of teaching math, coding, common sense reasoning, general knowledge of the world...
- •Acquired academic books and Q&A datasets
- •High quality chat format supervised data covering various topics to reflect human preferences on different aspects such as instruct-following,...
相关模型
快速推荐
最佳硬件
Phi-4 14B 的最佳选择
运行此模型
量化选项
各量化级别的 VRAM 估算
No hardware detected — fit column shows raw VRAM estimates
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 5.5 GB | Low | — |
Q3_K_S | 3 | 6.9 GB | Low | — |
NVFP4 | 4 | 7.8 GB | Medium | — |
Q4_K_M | 4 | 8.5 GB | Medium | — |
Q5_K_M | 5 | 10.1 GB | High | — |
Q6_K | 6 | 11.5 GB | High | — |
Q8_0 | 8 | 15.0 GB | Very High | — |
F16 | 16 | 28.7 GB | Maximum | — |
Quality benchmarks
Phi-4 14B benchmark scores
Coding
SWE-bench Verified—
HumanEval+82.6%
Aider Polyglot—
LiveCodeBench—
Reasoning
MMLU-Pro70.4%
GPQA Diamond56.1%
MATH-50080.4%
ARC Challenge—
General
Chatbot Arena—
IFEval63.0%
Source: official · 2024-12-12
硬件兼容性
全部硬件的适配估算
Computing compatibility...
内存详细分析
Reference: RTX 2060 6GB
Weights8.5 GB
KV Cache3.1 GB
Runtime1.2 GB
Headroom0.6 GB
常见问题
FAQ — Phi-4 14B
另请参阅