DeepSeek V4 — DeepSeek 的本地 AI 模型
作者： Jakub Rusinowski · 最后更新：
DeepSeek's April 24, 2026 preview release, MIT licensed with a 1M-token context window. V4-Pro (1.6T total / 49B active) is server-class only — not runnable on consumer hardware even heavily quantized. V4-Flash (284B total / 13B active) is workstation-tier but local support is still WIP: llama.cpp has no merged upstream support as of June 2026 (only experimental forks), and Ollama's listings appear to route to cloud-hosted inference rather than a true local download. DeepSeek R2 remains unreleased/rumored as of June 15, 2026 and is intentionally not listed.
变体
DeepSeek V4 最小的变体在 Q4 (experimental) 下约需 172 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 |
|---|---|---|---|---|
| DeepSeek V4-Flash → 284B (13B active) | ~172.3 GB | 1,000,000 | 云端托管标签 | |
| DeepSeek V4-Pro → 1.6T (49B active) | ~966.8 GB | 1,000,000 | ollama run deepseek-v4 |
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 DeepSeek V4
安装 Ollama，然后拉取标签。
Cloud-hosted on Ollama. Local inference is work-in-progress llama.cpp forks only.
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: DeepSeek V4-Flash, DeepSeek V4-Pro推荐 GPU
目录中能在本地运行 DeepSeek V4（至少 172 GB 显存）的最便宜 GPU 是 Apple M3 Ultra (512 GB).
我的 GPU 能运行 DeepSeek V4 吗？
- DeepSeek V4 在 AMD Ryzen AI Max+ 395 上
- DeepSeek V4 在 Apple M1 Ultra 上
- DeepSeek V4 在 Apple M2 Max 上
- DeepSeek V4 在 Apple M2 Ultra 上
- DeepSeek V4 在 Apple M4 Max 上
- DeepSeek V4 在 Apple M5 Max 上
- DeepSeek V4 在 NVIDIA A100 80GB (PCIe) 上
- DeepSeek V4 在 NVIDIA DGX Spark 上
- DeepSeek V4 在 NVIDIA H100 80GB (PCIe) 上
- DeepSeek V4 在 NVIDIA RTX PRO 6000 Blackwell 上
DeepSeek V4 — 常见问题
How much VRAM does DeepSeek V4 need?
DeepSeek V4 needs about 172 GB VRAM at Q4 (experimental) quantization for its smallest variant. Variants: DeepSeek V4-Flash (172 GB, Q4 (experimental)); DeepSeek V4-Pro (967 GB, Q2 (experimental, datacenter only)). On Apple Silicon, unified memory counts toward this requirement.
Can I run DeepSeek V4 on an RTX 4090 (24 GB)?
DeepSeek V4's smallest variant needs about 172 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 DeepSeek V4?
Q4_K_M is the best balance of quality and VRAM for DeepSeek V4 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 DeepSeek V4 with Ollama?
DeepSeek V4 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.