vllm-project/recipes

vLLM Recipes is a collection of community-maintained guides for running vLLM with specific models on particular hardware configurations.

The project addresses the practical challenge of deploying large language models by providing step-by-step recipes that answer the question: how do I run model X on hardware Y for task Z? Rather than offering a single unified approach, it curates model-specific and hardware-specific guidance contributed by the community, allowing users to find deployment patterns tailored to their exact combination of model and infrastructure.

Developers should adopt this resource when they need concrete deployment instructions for a specific model-hardware pairing. The collection covers a wide range of models from providers including DeepSeek, Google, Llama, Mistral AI, and others, making it valuable for teams working with any of these models. The guide-based approach works best for users who want to move beyond generic documentation and learn from configurations that others have already tested and validated.

The project maintains an active collection of model-specific recipes organized by provider, with guides covering recent model releases alongside established ones. Community contributions drive the expansion of available recipes, indicating ongoing engagement with the resource as new models and deployment scenarios emerge.