The easiest way to have no-downtime upgrades is have an architecture that can tolerate some subset of their processes to be down at any time. De-SPOF and this gets easier (not that de-SPOFing is always trivial).
Recent articles
- Gemini 2.0 Flash: An outstanding multi-modal LLM with a sci-fi streaming mode - 11th December 2024
- ChatGPT Canvas can make API requests now, but it's complicated - 10th December 2024
- I can now run a GPT-4 class model on my laptop - 9th December 2024