GLM-4.6V-Flash 9B — 显存、速度与本地部署
作者: Jakub Rusinowski · 最后更新:
The 9B vision-language model in the GLM-4.6V family, built for low-latency multimodal work. The real small GLM: GLM-4.7 is a ~400B MoE and has no 9B checkpoint.
GLM-4.6V-Flash 9B 在 Q4_K_M 下约需 6 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)。仅含权重与开销:该模型架构未公开,因此未计入 KV 缓存。
| 量化 | 显存 | 显存 | 速度(估算) | 适配 |
|---|---|---|---|---|
| Q2_K 2.63 bpw | 3.8 GB | ~144 tok/s | 可运行 | |
| Q3_K_M 3.41 bpw | 4.6 GB | ~123 tok/s | 可运行 | |
| Q4_K_M 4.83 bpw | 6.2 GB | ~98 tok/s | 可运行 | |
| Q5_K_M 5.67 bpw | 7.2 GB | ~87 tok/s | 可运行 | |
| Q6_K 6.56 bpw | 8.2 GB | ~78 tok/s | 可运行 | |
| Q8_0 8.50 bpw | 10.4 GB | ~64 tok/s | 可运行 | |
| F16 16.00 bpw | 18.8 GB | ~37 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. GLM-4.6V-Flash 9B 显存计算器 →
运行 GLM-4.6V-Flash 9B
目录中能运行 GLM-4.6V-Flash 9B 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 GLM-4.6V-Flash 9B
安装 Ollama,然后运行:
ollama run glm-4-6v规格
Corroborated — Two or more independent sources agree on these figures, but the model card itself was not retrieved. Treat the numbers as good rather than confirmed. 尚未确认: context.
- 参数量
- 9 Billion
- 上下文窗口
- 65,536
- 架构
- Dense vision-language
- 提供商
- Zhipu AI (Z.ai)
- 许可证
- MIT
- 规格量化
- Q4_K_M
- 系统内存
- 16 GB
- 记录更新于
- 2026-09-19
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。
GLM-4.6V-Flash 9B — 常见问题
How much VRAM does GLM-4.6V-Flash 9B need?
About 6 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 GLM-4.6V-Flash 9B run on an RTX 4090 (24 GB)?
Yes. GLM-4.6V-Flash 9B needs about 6 GB at Q4_K_M, inside a 24 GB card, at an estimated 98 tokens/sec.
How do I run GLM-4.6V-Flash 9B locally?
Install Ollama and run `ollama run glm-4-6v`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.