Run DeepSeek-Coder-33B Instruct, a code model trained on 2 trillion tokens, behind an OpenAI-compatible API in your own AWS account. Keep your source code off third-party AI services.
DeepSeek Coder is a series of code language models trained from scratch on 2 trillion tokens, 87% code and 13% natural language in English and Chinese, across 80 programming languages. The 33B Instruct model is fine-tuned on 2 billion instruction tokens.
Project-level completion with a 16K token window.
Infill missing code between existing lines.
Trained across a wide range of programming languages.
Leading results on HumanEval, MultiPL-E, MBPP, DS-1000 and APPS.
Completions, embeddings and model listing endpoints.
Your code stays in your own AWS account.
The AMI ships DeepSeek-Coder-33B Instruct and an OpenAI-compatible API server on Ubuntu 22.04.
Make sure your account has vCPU quota for g5g instances in your region.
Subscribe on AWS Marketplace and launch the CloudFormation stack with your domain and admin email.
After 5-10 minutes, create a DNS record with the PublicIp from the stack outputs.
Open /docs on your server, then use the completions endpoint from your tools.
An AMI on AWS Marketplace, launched on a g5g GPU instance in your own AWS account with a CloudFormation stack.
Launch on AWS | Endpoint | Purpose |
|---|---|
| /v1/completions | Code completions from a prompt |
| /v1/embeddings | Embeddings for input text |
| /v1/models | List available models |
| /v1/chat/completions | Available, but avoid: the model is not optimised for chat |
Use /v1/completions for coding work. Read the DeepSeek Coder developer guide →
A series of code language models trained on 87% code and 13% natural language, from 1.3B to 33B parameters. This AMI runs the 33B Instruct model.
The model is not optimised for chat, so avoid /v1/chat/completions and use /v1/completions and /v1/embeddings instead.
The AWS listing recommends g5g.16xlarge or g5g.metal.
16,000 tokens, which allows project-level completion and infilling.
Open /docs on your server, or use the Node.js test script from the developer guide.
SSH in and run sudo /root/certificate_generate_standalone.sh with your admin email.
Back up your server data, remove the old stack and launch the new version from AWS Marketplace.
The open-source self-hosted AI coding model, deployed with Meetrix.
Where DeepSeek fits among open models.
Launch the stack, set up SSL and test the API.
We build and run self-hosted AI infrastructure every day, from coding assistants to GPU sizing. Tell us what you need.
Contact us