这两天,DeepSeek接连发布了升级版R1模型DeepSeek-R1-0528及蒸馏模型DeepSeek-R1-0528-Qwen3-8B,模型性能相较于此前R1模型大幅升级。DeepSeek-R1-0528性能自是不必多言,值得注意的是,根据升级版R1模型思维链提取出来蒸馏模型性能依旧强大,在数学测试中表现与 Qwen3-235B 相当,可以说是"小参数大性能"。
此前,不少企业和个人尝试将模型进行私有化部署,但是,如果将DeepSeek-R1(671B)进行私有化部署,至少需要1.5TB显存,按官方推荐配置需要将近20张NVIDIA H100 80GB显卡,门槛较高。
而小参数的DeepSeek-R1-0528-Qwen3-8B,则可以完美匹配这一需求。该模型在保持较高推理能力的同时大幅降低了运行资源需求,适合个人及中小企业部署。现在PPIO派欧云上线了该模型私有化部署模板,用户可一键将DeepSeek-R1-0528-Qwen3-8B部署在云服务器上,10分钟就能拥有专属模型。
模型私有化部署教程
选择模板与配置
打开PPIO派欧云算力市场,选择 deepseek-ai/DeepSeek-R1-0528-Qwen3-8B 模版,选择卡型 4090 点击部署。
算力市场地址:ppio.cn/gpu-instanc...

确认部署参数
检查部署参数确认页面。确认无误后,点击 下一步
。

启动部署
- 在最终确认页面,点击
部署
按钮启动实例创建。 此过程可能需要几分钟,请耐心等待。

查看部署进度
点击实例管理,进入控制台查看部署进度


点击实例,查看镜像拉取进度

等待实例变为运行中,点击日志 ---> 实例日志查看模型拉取进度,等待模型拉取完成


日志出现 "Application startup complete." 即为部署完成!

获取模型访问地址
点击实例连接,拷贝端口 8000 的 HTTP 映射,即模型访问

访问您的私有模型
注意请将 下文中的 " c9f957427e53e9ba-8000.cn-northwest-2.gpu-instance.ppinfra.com " 替换为您真正的访问地址,复制以下代码,访问您的私有模型!
swift
# Call the server using curl:
curl -X POST "https://c9f957427e53e9ba-8000.cn-northwest-2.gpu-instance.ppinfra.com/v1/chat/completions" \
-H "Content-Type: application/json" \
--data '{
"model": "deepseek-ai/DeepSeek-R1-0528-Qwen3-8B",
"messages": [
{
"role": "user",
"content": "What is the capital of France?"
}
]
}'
{"id":"chatcmpl-d9e5f06123944037a8105d638d902f06","object":"chat.completion","created":1748594949,"model":"deepseek-ai/DeepSeek-R1-0528-Qwen3-8B","choices":[{"index":0,"message":{"role":"assistant","reasoning_content":null,"content":"<think>\nOkay, the user is asking for the capital of France. This seems like a straightforward factual question, probably aimed at getting a quick answer. \n\nThe user might be a student doing homework, someone preparing for a quiz, or just an individual curious about geography. The question's simplicity suggests they don't necessarily need extra context, just the direct answer. \n\nI recall that Paris is indeed France's capital, with a long history dating back centuries. There's no ambiguity here---it's a well-established fact. The response should be concise to match the query's straightforward nature.\n\nConsidering possible follow-up questions, I might mentally prepare to elaborate if the user asks about Paris's history or significance, but since the query is direct, there's no need to overload with information right now. \n\nThe answer "Paris" is correct and unambiguous. No need to complicate things unless the user indicates further interest.\n</think>\nThe capital of France is **Paris**.","tool_calls":[]},"logprobs":null,"finish_reason":"stop","stop_reason":null}],"usage":{"prompt_tokens":10,"total_tokens":204,"completion_tokens":194,"prompt_tokens_details":null},"prompt_logprobs":null}

将 API 地址配置到您的 chatbox 等应用,就可以拥有专属模型及聊天助手!
目前,PPIO算力市场已上线18个私有化部署模板,除了DeepSeek-R1-0528-Qwen3-8B,用户也可以将llama3-8b:v0.5.0 、StableDiffusion:v1.8.0等模型快速进行私有化部署,帮助企业及个人降低模型部署成本,实现高效、安全调用。