Configuration Examples

These examples cover common service topologies. Model paths, GPU assignments, profile aliases, and memory budgets depend on the deployment environment.

File	Topology
simple.yaml	One node, 1 Simple Engine, TP=1
pd.yaml	One node, 1P+1D, TP=1
multi-node.yaml	Two physical nodes, 1 Simple Engine per node
dynamic.yaml	Empty startup followed by explicit deployment
work-stealing.yaml	One node, 1P+2D
foundry-save.yaml / foundry-load.yaml	P/D Decode, batch=1, one 4096 bucket; requires Foundry and the hook; SAVE creates archives, LOAD requires complete archives
chat-request.json	Minimal nonstreaming greedy request

curl -sS http://127.0.0.1:8000/v1/chat/completions -H 'Content-Type: application/json' --data-binary @docs/en/examples/chat-request.json

Concurrent services need disjoint GPU assignments and separate shared-memory directories. SAVE and LOAD archives require matching layouts, models, and dependencies.