devtool
Kvcachescope – Why Nvidia-smi is blind to vLLM KV cache leaks
It can spot vLLM KV cache leaks, but its narrow scope and shaky maintenance make it unsuitable as a production tool.
4.8Overall
Utility5
Onboarding5
Craft5
Niche fit6
Longevity3
Five dimensions scored independently; overall is a weighted average. Scores are only comparable within this same rubric.
Good for
Researchers or ML Ops engineers debugging vLLM inference
Not for
Production teams requiring stable, well‑maintained tooling
Project description
github.com
Alternatives
nvidia-smivLLM built‑in profilingtorch.cuda.memory_summary