Deep Seek 🐋

A simple wrapper around DeepSeek which exposes an API via Bun and Docker.

Prerequisites

NOTE: The Docker commands will start a container running Ollama and then download the DeepSeek model. This container exposes the API on port 11434 and the optional Bun server will serve a simple we chat interface at http://localhost:3000.

• Docker Desktop
• Bun (Optional)

Quick Start

# Start the docker container (first time running may take longer)
docker compose up -d

# List all available models (via Docker Desktop CLI)
ollama list

# If you don't see the DeepSeek model you may need to manually pull it
# by running the following in the Docker Desktop CLI:
ollama pull deepseek-r1:7b

# Quickly test the model is running (via Docker Desktop CLI)
curl http://localhost:11434/api/generate -d '{
  "model": "deepseek-r1:7b",
  "prompt": "Please tell me an interesting factoid?"
}'

# (Optional) Start the Bun server to interact with web chat,
# run outside the container to chat on http://localhost:3000/
bun run dev

# Stop the container
docker compose down

DeepSeek Models

• https://ollama.com/library/deepseek-r1

# DeepSeek Models
ollama run deepseek-r1:1.5b
ollama run deepseek-r1:7b       # Default
ollama run deepseek-r1:8b
ollama run deepseek-r1:14b
ollama run deepseek-r1:32b
ollama run deepseek-r1:70b
ollama run deepseek-r1:671b     # o1 Level

Installation

# If this is your first time build the Docker container,
# you may need to manually increase the memory to 8.8GB+
bun run docker:build
bun run docker:run

# You can test it's running by executing the following:
curl http://localhost:11434/api/generate -d '{
  "model": "deepseek-r1:7b",
  "prompt": "Hello!"
}'

# NOTE: To run this program, do the following:
bun install
bun run dev

Ollama Commands

The following are useful commands for interacting with the Ollama CLI:

# List all available models
ollama list

# Download a model
ollama pull deepseek-r1:1.5b

# Run a model
ollama run deepseek-r1:1.5b

Project Structure

This project is faily minimal and contains some code for the Bun server which returns some basic frontend code for a chat interface.

• ./docker-compose.yml - Docker compose file for running the DeepSeek container.
• ./index.ts - Bun entry point
• ./src/* - Root project directory.
• ./src/server/* - Contains the Bun server-side code.
• ./src/client/* - Contains the HTML, CSS & JS client-side code for chat.
• ./src/client/index.html - The main HTML file for the chat interface.
• ./src/client/assets/style.css - The CSS file for the chat interface.
• ./src/client/assets/script.js - The JS file for the chat interface. (Client logic)

The UI is just basic HTML, CSS, and JS and is really only meant to help test the model.

NOTE: the chat doesn't support markdown, LaTeX or other fancy things at the moment.

Ollama Params

The following example type represents the request body for the Llama2 model, although it seems all models on Ollama support these params, however, the deep seek team says the latest model may not work as expected with items like context or system.

type OllamaRequest = {
  "model": "deepseek-r1:7b",// Model name
  "prompt": "string",       // Input prompt
  "stream": boolean,        // Stream responses
  "raw": boolean,           // Raw model output
  "template": "string",     // Custom prompt template
  "context": [],            // Previous context array
  "system": "string",       // System prompt
  "temperature": float,     // 0-1, default 0.8
  "top_p": float,           // 0-1, default 0.9
  "top_k": int,             // 1-100, default 40
  "num_predict": int,       // Max tokens to generate
  "stop": ["string"],       // Stop sequences
  "repeat_penalty": float   // 0-2, default 1.1
}

Additional Resources

• DeepSeek-R1 GitHub Page
• Ollama Website