vLLM issue #2248: recent vLLMs ask for too much memory, ValueError: No available memory for the cache blocks

Long-running issue (closed 2025-08-14) collecting several distinct reports about KV cache block allocation failing on small models and small max-model-len. The --gpu_memory_utilization semantics are the common thread; the individual configurations do not share one mechanism. Confidence is set below the other gotcha sources because the thread is an aggregation, not a single root-caused repro.