Tag: Inference Infrastructure

2 articles

  • This article covers production governance for LoRA adapter caches in multi-tenant LLM serving, including dynamic loading, hot/cold tiering, eviction strategies, tenant quotas, version releases, and go-live checks to reduce first-token latency and GPU memory jitter.

  • A practical guide to building overload protection for LLM inference services using SLO-driven autoscaling and admission control. Covers queue length, token pressure, TTFT/TPOT, and tenant quotas to prevent system collapse under traffic spikes.