SmolLM2 360M Instruct — 显存、速度与本地部署
作者： Jakub Rusinowski · 最后更新：
Incredibly lightweight. Runs on any device including old phones and embedded systems. Useful for on-device classification, simple Q&A, and keyword extraction where even 1B is too large.
SmolLM2 360M Instruct 在 Q4_K_M 下约需 1 GB 显存——量化权重加框架开销，不含 KV 缓存。在 Apple Silicon 上，这部分来自统一内存。
按量化级别的显存与速度
计算基准：NVIDIA RTX 4090 (24 GB)。包含 8K 上下文的 KV 缓存，因此数值高于上方的主数字。
| 量化 | 显存 | 显存 | 速度（估算） | 适配 |
|---|---|---|---|---|
| Q2_K 2.63 bpw | 1.3 GB | ~381 tok/s | 可运行 | |
| Q3_K_M 3.41 bpw | 1.3 GB | ~374 tok/s | 可运行 | |
| Q4_K_M 4.83 bpw | 1.4 GB | ~363 tok/s | 可运行 | |
| Q5_K_M 5.67 bpw | 1.4 GB | ~356 tok/s | 可运行 | |
| Q6_K 6.56 bpw | 1.4 GB | ~350 tok/s | 可运行 | |
| Q8_0 8.50 bpw | 1.5 GB | ~336 tok/s | 可运行 | |
| F16 16.00 bpw | 1.9 GB | ~292 tok/s | 可运行 |
黑色标记 = NVIDIA RTX 4090 (24 GB) 上的可用显存。 估算来自内存带宽屋顶线模型，详见 方法说明页. SmolLM2 360M Instruct 显存计算器 →
运行 SmolLM2 360M Instruct
目录中能运行 SmolLM2 360M Instruct 的最便宜 GPU 是 Intel Arc B570 (10 GB).
或在 Vast.ai 比较，低至 $0.35/小时 (typical low · varies)
作为亚马逊联盟成员，我们从符合条件的购买中获得收入。云 GPU 链接为推荐链接——我们可能获得佣金，您无需额外付费。
如何运行 SmolLM2 360M Instruct
安装 Ollama，然后运行：
ollama run smollm2:360m规格
Curated — A hand-written entry from before this catalogue recorded its sources. The figures are long-standing but their provenance is not on file.
- 参数量
- 360 Million
- 上下文窗口
- 8,192
- 架构
- Dense
- 提供商
- Hugging Face
- 许可证
- Apache 2.0
- Practical deployment
- Edge device (phone, SBC)
- 规格量化
- Q4_K_M
- 系统内存
- 2 GB
- 记录更新于
- 2024-11-20
Commercial use permitted. No usage restrictions beyond attribution.
质量与使用场景
评分由模型作者或独立评测方发布——衡量质量而非吞吐量，并非我们实测。
我的 GPU 能运行 SmolLM2 360M Instruct 吗？
SmolLM2 的其他尺寸
SmolLM2 360M Instruct — 常见问题
How much VRAM does SmolLM2 360M Instruct need?
About 1 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.
Does SmolLM2 360M Instruct run on an RTX 4090 (24 GB)?
Yes. SmolLM2 360M Instruct needs about 1 GB at Q4_K_M, inside a 24 GB card, at an estimated 363 tokens/sec.
How do I run SmolLM2 360M Instruct locally?
Install Ollama and run `ollama run smollm2:360m`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.
What other sizes does SmolLM2 come in?
SmolLM2 1.7B Instruct (2 GB), SmolLM2 360M Instruct (1 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.