StepFun Step — StepFun 的本地 AI 模型
作者： Jakub Rusinowski · 最后更新：
阶跃星辰（StepFun）的开放权重混合专家模型，交替使用滑动窗口注意力和全注意力，使长上下文缓存保持较小：Step 3.7 Flash（Apache-2.0）和 600B 的 Step 5 Preview（StepFun Model License 1.0）。
变体
StepFun Step 最小的变体在 Q4_K_S 下约需 112 GB 显存——量化权重加框架开销，不含 KV 缓存。
| 模型 | Q4 下显存 | 显存 | 上下文 | 运行 |
|---|---|---|---|---|
| Step 3.7 Flash 198B-A11B → 198B (11B active) | ~112.3 GB | 262,144 | 云端托管标签 | |
| Step 5 Preview 600B-A27B → 600B (27B active) | ~365.6 GB | 1,048,576 | 云端托管标签 |
显存为 Q4_K_M 下的量化权重加开销，与 GPU 与显存检测器使用同一引擎计算。
如何在本地运行 StepFun Step
安装 Ollama，然后拉取标签。
No Ollama tag. Use StepFun's official GGUFs with llama.cpp (mainline supports the Step 3.5 architecture the model uses; the card's build steps use StepFun's own fork), or vLLM / SGLang.
在上方选择一个尺寸，查看它自己的显存、速度估算和安装命令。
许可证
Commercial use permitted. No usage restrictions beyond attribution.
适用于: Step 3.7 Flash 198B-A11BWeights are downloadable and commercial use is permitted, subject to the licence’s acceptable-use terms.
适用于: Step 5 Preview 600B-A27B推荐 GPU
目录中能在本地运行 StepFun Step（至少 112 GB 显存）的最便宜 GPU 是 NVIDIA DGX Spark (128 GB).
StepFun Step — 常见问题
How much VRAM does StepFun Step need?
StepFun Step needs about 112 GB VRAM at Q4_K_S quantization for its smallest variant. Variants: Step 3.7 Flash 198B-A11B (112 GB, Q4_K_S); Step 5 Preview 600B-A27B (366 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
Can I run StepFun Step on an RTX 4090 (24 GB)?
StepFun Step's smallest variant needs about 112 GB, which exceeds a single RTX 4090 (24 GB). Use multiple GPUs, a higher-VRAM card, or Apple Silicon with large unified memory.
What quantization should I use for StepFun Step?
Q4_K_M is the best balance of quality and VRAM for StepFun Step in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
How do I run StepFun Step with Ollama?
StepFun Step has no local Ollama tag — the published tag is cloud-hosted, so running it sends your prompts to a hosted GPU rather than your own machine.