Back to platforms

vLLM Architecture Lab

High-throughput LLM inference — PagedAttention, batching, KV cache.

Educational vLLM reference: PagedAttention, continuous batching, KV budgets, multi-LoRA economics — pairs with DomainForge adapter training (ADR-022).

In one sentence

Interactive vLLM architecture explorer — PagedAttention, batching, KV budget, adapter chat.

Decision

Educational simulator before GPU spend — teams learn inference architecture without H100 budget.

Measured signal

Live simulator · KV budget calculator · DomainForge↔vLLM Path B (ADR-022 Accepted).

Honest limitation

Simulator + educational adapter swap — not CUDA multi-LoRA kernels or production SLAs.

  • PagedAttention block allocator with swap and prefix sharing
  • FCFS scheduler — waiting / running / swapped queues
  • KV memory budget calculator (H100 + Llama-3)
  • Live API on Render · architecture explorer on Vercel
  • Educational multi-LoRA Path B — `/v1/chat/completions` + `/v1/adapters`

Read related ADR →

Multi-agent reference topologyArchitecture
EXPERIENCECONTROL PLANEMODEL PLANEOBSERVABILITYUser / OpsPolicy & GuardrailsAegisAIOrchestratorLangGraphSpecialist AgentsHybrid RAGTools / APIsEvaluation GatesGateway / HITLLLM Gatewayaegis-llm-gatewaySemantic Cacheaegis-semantic-cacheTraces · Audit · FinOpstrace-linked LLMOps

Python · FastAPI · Vercel · Render