DeepSeek R1 & Local LLMs: Running, Fine-Tuning & Deploying with Ollama

By Elena Rostova • Intermediate • 15 min read • Updated 2026-09-12

What You Will Master in This Tutorial

Understand Reinforcement Learning for reasoning models (DeepSeek R1).
Run quantized GGUF models on consumer GPUs using Ollama.
Integrate local LLM endpoints into Python microservices via the OpenAI SDK.
Deploy vLLM in Docker containers for production enterprise inference.

1. Running DeepSeek R1 with Ollama in Seconds

DeepSeek R1 revolutionized open-weights AI by demonstrating that reinforcement learning reasoning can match closed frontier models. With Ollama, you can download and run quantized versions directly on your workstation.

BASH

# Download and start DeepSeek R1 distilled 8B model
ollama run deepseek-r1:8b

# Or run the larger 14B / 32B model on high-VRAM machines
ollama run deepseek-r1:14b

Note: An 8B model requires only ~6GB of VRAM and runs smoothly on Apple Silicon M-series or NVIDIA RTX GPUs.

2. Querying Ollama via Python and Streaming Responses

Ollama exposes an OpenAI-compatible HTTP API at http://localhost:11434/v1, allowing drop-in replacement with standard client SDKs.

PYTHON

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:11434/v1",
    api_key="ollama"  # Required but ignored locally
)

response = client.chat.completions.create(
    model="deepseek-r1:8b",
    messages=[
        {"role": "system", "content": "You are a C# and Python expert."},
        {"role": "user", "content": "Explain how virtual threads differ from async/await."}
    ],
    stream=True
)

for chunk in response:
    print(chunk.choices[0].delta.content or "", end="", flush=True)

Note: Using local models guarantees 100% data privacy and zero API token costs.

Knowledge Check: Test Your Understanding

1. What is the primary breakthrough introduced by DeepSeek R1?

Frequently Asked Questions

Can I connect Ollama to Cursor or VS Code?

Yes! Extensions like Continue.dev, Roo Code, and Cline allow using http://localhost:11434 as your primary coding backend.