全系列 DeepSeek R1 模型单机部署配置要求

GrissomFI2025年02月14日约 2 分钟

单机部署要求

注意是单机，不是集群！

参考配置如下：

版本	名称	参数（亿个）	数据类型	模型大小（G）	模型格式	发布者	单机部署要求
满血版	DeepSeek-R1	6710	BF8	700	safetensors	DeepSeek	8 * Nvidia H200 (141G)
8 * AMD MI300X (192G)
满血量化版	deepseek-r1:671b	6710	INT4	404	GGUF	Ollama	8 * Nvidia H100 (80G)
8 * 昇腾 910B (64G)
蒸馏版	DeepSeek-R1-Distill-Llama-70B	700	BF16	150	safetensors	DeepSeek	8 * 3090 (24G)
DeepSeek-R1-Distill-Qwen-32B	320	66	DeepSeek	8 * 2080Ti (11G)
DeepSeek-R1-Distill-Qwen-14B	140	30	DeepSeek	2 * 2080Ti (11G)
DeepSeek-R1-Distill-Llama-8B	80	16	DeepSeek	PC
DeepSeek-R1-Distill-Qwen-7B	70	15	DeepSeek	PC
DeepSeek-R1-Distill-Qwen-1.5B	15	3.5	DeepSeek	移动端、边缘计算、PC
蒸馏量化版	deepseek-r1:70b	700	INT4	43	GGUF	Ollama	2 * 3090 (24G)
deepseek-r1:32b	320	20	Ollama	PC
deepseek-r1:14b	140	9	Ollama	PC
deepseek-r1:8b	80	4.9	Ollama	移动端或PC
deepseek-r1:7b	70	4.7	Ollama	移动端或PC
deepseek-r1:1.5b	15	1.1	Ollama	移动端、边缘计算

满血版部署参考

1. Deploy DeepSeek-R1 671B on 8x NVIDIA H200 with SGLang：https://datacrunch.io/blog/deploy-deepseek-r1-on-8x-nvidia-h200
2. Deploy Deepseek-R1 in one GPU -AMD Instinct™ MI300X：https://medium.com/@alexhe.amd/deploy-deepseek-r1-in-one-gpu-amd-instinct-mi300x-7a9abeb85f78
3. 4090单卡跑满血版DeepSeek-R1，清华团队开源项目再破大模型推理门槛

最新官方技巧

🎉 Excited to see everyone’s enthusiasm for deploying DeepSeek-R1! Here are our recommended settings for the best experience:

• No system prompt
• Temperature: 0.6
• Official prompts for search & file upload: http://bit.ly/4hyH8np
• Guidelines to mitigate model bypass thinking: http://bit.ly/4gJrhkF

The official DeepSeek deployment runs the same model as the open-source version—enjoy the full DeepSeek-R1 experience! 🚀