Running open large language models in production with Ollama and serverless GPUs by Wietse Venema