Estimate the KV cache for a model you choose and check it against measured memory

Compute KV bytes per token from a model configuration, predict memory for several context lengths and batch sizes, and compare with measured allocations.

The exercise happens on Self-directed (Hugging Face Transformers or vLLM). Canonical link: https://huggingface.co/docs/transformers/en/kv_cache. Link check: link responded (5 Oct 2026). Opened from this app 0 times; opening is not counted as an attempt.

Supports: Needs a GPU that fits a small open model. Prepares quick KV-cache arithmetic for inference interviews.