Meta
Llama 4 Scout 17B 16E
前沿
425.5K下载量1.3K点赞Apr 2025发布日期10.5M tokens上下文Llama 4 Community许可证66 良好质量
Llama 4 Scout 17B 16E (109B parameters) requires approximately 71.2 GB of VRAM with Q4_K_M quantization. As a Mixture of Experts model with 17B active parameters, it uses less memory than its total parameter count suggests. For the best balance of quality and speed, we recommend hardware with at least 82 GB of VRAM.
快速开始
— 复制粘贴即可本地运行Copy-paste commands to run Llama 4 Scout 17B 16E on your machine.
Run
lms load Llama-4-Scout-17B-16E-Instruct && lms server startQuick specs
Parameters109B (17B active)
Architecturemoe (MoE)
Context10.5M tokens
Modalitytext+vision
Min RAM42.5 GB
Rec. RAM66.5 GB (Q4_K_M)
LicenseLlama 4 Community
FamilyLlama
✓ Vision✓ Chat✓ Reasoning
About this model
相关模型
快速推荐
最佳硬件
Llama 4 Scout 17B 16E 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 | 42.5 GB | Low | — |
Q3_K_S | 3 | 53.4 GB | Low | — |
NVFP4 | 4 | 61.0 GB | Medium | — |
Q4_K_M | 4 | 66.5 GB | Medium | — |
Q5_K_M | 5 | 78.5 GB | High | — |
Q6_K | 6 | 89.4 GB | High | — |
Q8_0 | 8 | 116.6 GB | Very High | — |
F16 | 16 | 223.5 GB | Maximum | — |
Quality benchmarks
Llama 4 Scout 17B 16E benchmark scores
Coding
SWE-bench Verified—
HumanEval+—
Aider Polyglot—
LiveCodeBench32.8%
Reasoning
MMLU-Pro74.3%
GPQA Diamond57.2%
MATH-50050.3%
ARC Challenge—
Source: official · 2025-04-05
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
Weights66.5 GB
KV Cache2.9 GB
Runtime1.2 GB
Headroom0.6 GB
常见问题
FAQ — Llama 4 Scout 17B 16E
另请参阅