_CORE
AI & Agentic Systems Core Information Systems Cloud & Platform Engineering Data Platform & Integration Security & Compliance QA, Testing & Observability IoT, Automation & Robotics Mobile & Digital Banking & Finance Insurance Public Administration Defense & Security Healthcare Energy & Utilities Telco & Media Manufacturing Logistics & E-commerce Retail & Loyalty
References Technologies Blog Know-how Tools
About Collaboration Careers
CS EN
Let's talk

Ollama vs vLLM

14. 03. 2024 1 min read intermediate

Ollama is the simplest path to local LLMs. vLLM is optimized for production serving.

Ollama

  • Simple installation (curl + ollama run)
  • Model management (pull, list, rm)
  • REST API compatible with OpenAI
  • Ideal for development and experimentation
  • macOS, Linux, Windows

ollama pull llama3.2 ollama run llama3.2 ‘Explain Docker’ curl http://localhost:11434/api/generate -d ‘{“model”:”llama3.2”,”prompt”:”Hello”}’

vLLM

  • PagedAttention — efficient GPU memory management
  • Continuous batching — high throughput
  • OpenAI-compatible API server
  • Tensor parallelism (multi-GPU)
  • Optimized for production

pip install vllm python -m vllm.entrypoints.openai.api_server \ –model meta-llama/Llama-3-8B-Instruct

Comparison

  • Simplicity: Ollama >> vLLM
  • Throughput: vLLM >> Ollama (2-5×)
  • GPU utilization: vLLM better
  • Model format: Ollama = GGUF, vLLM = HuggingFace
  • CPU inference: Ollama OK, vLLM GPU-only

Ollama for Dev, vLLM for Production

Ollama for local development and experimentation. vLLM for production serving with high throughput.

ollamavllmllmaiinference
Share:

CORE SYSTEMS tým

Stavíme core systémy a AI agenty, které drží provoz. 15 let zkušeností s enterprise IT.