Self-host Deepseek R1 on AWS, GCP, Azure & K8s in Three Easy Steps

If you've been searching for a way to run your own large language model (LLM) without sacrificing performance, privacy, or cost-effectiveness, look no further than DeepSeek R1. This cutting-edge LLM solution enables you to deploy on spot-priced A100s and H100s in your own cloud infrastructure - Amazon Web Services, Google Cloud Platform, or Microsoft Azure - using Northflank's "Bring Your Own Cloud" (BYOC) feature. In under an hour, you can harness the power of DeepSeek while keeping your data entirely within your own cloud account.

DeepSeek R1 captures the zeitgeist of modern AI: it's fast, flexible, and ready to build on for production. Below is a quick walkthrough on how to self-host DeepSeek in your own cloud account using Northflank. Remove the complexity of Kubernetes while retaining control over your infrastructure and data.

Why Self-Host DeepSeek R1?

1. Complete Data Privacy: Maintain full control of your chat history, logs, and any uploaded data.
2. Lower Costs with Spot Instances: DeepSeek R1 can leverage spot-priced GPUs - like A100s and H100s - to deliver massive compute power without draining your budget.
3. Fast Setup: Get up and running in less than an hour - no advanced DevOps expertise needed.
4. Flexibility Across Clouds: Use AWS, GCP, or Azure - whichever platform works best for you.

Step 1: Prepare Your Cloud Provider and Northflank Account

1. Create/Log In to Your Cloud Provider Account

   • Set up a new project or resource group in AWS, GCP, or Azure.
   • Make sure your account has permissions to spin up GPU-based VMs or container instances (e.g., A100, H100, or whatever GPU you prefer).

2. Sign Up for Northflank

   • Create a Northflank account, then enable the BYOC functionality by linking your cloud provider credentials.

3. Check Your Cloud Quotas

   • Before deploying, ensure you have sufficient quota for the GPU resources you intend to use. Spot instances are cheaper but can be reclaimed by the cloud provider, so you'll want to plan for that.

Step 2: Deploy DeepSeek R1 via Northflank BYOC

1. Deploy DeepSeek from a Northflank stack template

   • Northflank offers templates which allow you to deploy DeepSeek into your chosen cloud provider with a couple of clicks. Click Deploy Deepseek Now for GCP, AWS, or Azure.

2. Configure the DeepSeek template

   • Select the integration for your chosen cloud provider. This ensures that all compute and storage come from your own cloud account.
   • Review the cluster configuration and resources that the template will create. You can proceed with the defaults, or select a different region or node types for your new cluster.
   • Deploy the stack to save the template in your team.

3. Run the DeepSeek template

   • Run the template and Northflank will provision a new cluster in your cloud account. When that's done, it'll create a new project and deploy the DeepSeek resources.

Step 3: Configure & Test Your LLM

1. Access DeepSeek via Open-WebUI

   • Once deployment finishes, navigate to the Open-WebUI endpoint and create a new account.

2. Run Sample Queries

   • Configure how you want DeepSeek R1 to respond to queries---tailor it to your business logic, data, or unique domain knowledge.
   • Test the LLM with a few prompts to ensure it's functioning as intended.
   • Fine-tune any parameters for latency, output style, or memory constraints.

3. Expand, Secure, and Iterate

   • Add more GPU nodes if you need additional throughput, or deploy nodes with more GPUs.
   • Get ready for production and switch to on-demand instances.
   • Keep iterating: one of the best parts about self-hosting is you can adapt as quickly as your business demands.

Watch the Demo

Watch how Northflank simplifies DevOps. A new cluster is created in the linked cloud provider account, and the required GPU workload - DeepSeek on OIlama in this case - is also provisioned along with Open-WebUI and persistent volumes. The command to download and run the DeepSeek model is executed when the application is running, meaning it's ready to use immediately. What would normally be a hassle at best and a nightmare at worst - provisioning a Kubernetes cluster in your own cloud account with all the tools required for a developer to immediately begin deploying workloads - can be done in a click!

This demo shows the DeepSeek 70B stack template for Azure.

Conclusion

DeepSeek R1 brings the promise of cutting-edge AI directly into your own environment, allowing you to tap into spot-priced GPUs with complete confidence in your data's security. With Northflank BYOC, orchestration becomes a breeze - spend less time worrying about infrastructure and more time developing your product.

Spin up DeepSeek R1 in GCP, AWS, or Azure today using Northflank, and experience the perfect blend of performance, security, and cost-efficiency.