Component guide
vllm
High-throughput LLM serving for GPUs.
Atlas
in AtlasStack-Complete-Atlas-+-Onboarding-300
vllm
Licence: Apache-2.0 | Language: Python | Source: https://github.com/vllm-project/vllm
Role in the stack
High-throughput LLM serving
Why it's here
PagedAttention serving with the best throughput per GPU for open models. The default when you have real GPU capacity.
Using it
See the project README in the fetched source:
./fetch-sources.sh ./sources
tar xzf ./sources/vllm.tar.gz -C ./sourcesRunning it
Started by docker compose up -d on port 8000.
This component ships inside the Atlas
stack. See pricing →