Inkling Small — 显存、速度与本地部署
作者： Jakub Rusinowski · 最后更新：
Thinking Machines' smaller open model: a mixture-of-experts with 276B parameters (266B in the published files) and 12B active per token. It takes text, images and audio, writes text, and has a 1M-token context. Apache-2.0; vLLM, SGLang and TokenSpeed recipes on the card.
Inkling Small 在 Q4_K_M 下约需 161 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 | 88.6 GB | — | 放不下 | |
| Q3_K_M 3.41 bpw | 114.5 GB | — | 放不下 | |
| Q4_K_M 4.83 bpw | 161.8 GB | — | 放不下 | |
| Q5_K_M 5.67 bpw | 189.7 GB | — | 放不下 | |
| Q6_K 6.56 bpw | 219.3 GB | — | 放不下 | |
| Q8_0 8.50 bpw | 283.8 GB | — | 放不下 | |
| F16 16.00 bpw | 533.1 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. Inkling Small 显存计算器 →
运行 Inkling Small
目录中能运行 Inkling Small 的最便宜 GPU 是 Apple M3 Ultra (512 GB).
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 Inkling Small
安装 Ollama，然后运行：
ollama run inkling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
- 276 Billion (12B active)
- 上下文窗口
- 1,048,576
- 架构
- Mixture-of-Experts (256 experts, 6 active), 7 global + 35 sliding-window (512) layers, image and audio encoders
- 提供商
- Thinking Machines
- 许可证
- Apache-2.0
- Availability
- Weights downloadable
- Practical deployment
- Workstation (large unified memory or pro GPU)
- 规格量化
- Q4_K_M
- 系统内存
- 256 GB
- 记录更新于
- 2026-11-11
Commercial use permitted. No usage restrictions beyond attribution.
限制与注意事项
- The card gives 276B parameters; the published safetensors hold 266.0B, and memory here is sized from the files.
另请参阅 Inkling (NVFP4)
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 Inkling Small 吗？
Inkling 的其他尺寸
Inkling Small — 常见问题
How much VRAM does Inkling Small need?
About 161 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 Inkling Small run on an RTX 4090 (24 GB)?
No. Inkling Small needs about 161 GB at Q4_K_M, more than a single RTX 4090's 24 GB. It needs a larger card, several GPUs, or Apple Silicon with enough unified memory — or it runs with part of the weights offloaded to system RAM, which is much slower.
How do I run Inkling Small locally?
Install Ollama and run `ollama run inkling`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.
What other sizes does Inkling come in?
Inkling (NVFP4) (589 GB), Inkling (BF16) (589 GB), Inkling Small (161 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.