Jais 3 32B — 显存、速度与本地部署
作者： Jakub Rusinowski · 最后更新：
The large dense Jais 3: 32B-class (34.8B in the files), Arabic and English, a 512K-token context and reasoning-effort control. Apache-2.0; official GGUF (Q4_K_M 21.1 GB), FP8 and MLX builds.
Jais 3 32B 在 Q4_K_M 下约需 22 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 | 14.4 GB | ~53 tok/s | 可运行 | |
| Q3_K_M 3.41 bpw | 17.8 GB | ~43 tok/s | 可运行 | |
| Q4_K_M 4.83 bpw | 24 GB | ~5 tok/s | 需卸载 | |
| Q5_K_M 5.67 bpw | 27.6 GB | ~4 tok/s | 需卸载 | |
| Q6_K 6.56 bpw | 31.5 GB | ~4 tok/s | 需卸载 | |
| Q8_0 8.50 bpw | 39.9 GB | ~3 tok/s | 需卸载 | |
| F16 16.00 bpw | 72.5 GB | — | 放不下 |
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. Jais 3 32B 显存计算器 →
运行 Jais 3 32B
目录中能运行 Jais 3 32B 的最便宜 GPU 是 NVIDIA GeForce RTX 5090 (32 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 Jais 3 32B
安装 Ollama，然后运行：
ollama run jais-3规格
Verified — Checked against the primary source — the model card or the vendor spec page — and corroborated by a second independent source.
- 参数量
- 32 Billion (34.8B in the files)
- 上下文窗口
- 524,288
- 架构
- Dense transformer (K2 Horizon, GQA)
- 提供商
- Inception / MBZUAI Institute of Foundation Models
- 许可证
- Apache-2.0
- Availability
- Weights downloadable
- Practical deployment
- Single consumer GPU
- 规格量化
- Q4_K_M
- 系统内存
- 32 GB
- 记录更新于
- 2026-11-11
Commercial use permitted. No usage restrictions beyond attribution.
另请参阅 Jais 3 7B
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。
Jais 3 的其他尺寸
Jais 3 32B — 常见问题
How much VRAM does Jais 3 32B need?
About 22 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 Jais 3 32B run on an RTX 4090 (24 GB)?
Yes. Jais 3 32B needs about 22 GB at Q4_K_M, inside a 24 GB card, at an estimated 5 tokens/sec.
How do I run Jais 3 32B locally?
Install Ollama and run `ollama run jais-3`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.
What other sizes does Jais 3 come in?
Jais 3 7B (6 GB), Jais 3 32B (22 GB), Jais 3 375B-A23B (230 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.