Context and KV cache

Inferlets that exercise shared KV prefixes and bounded attention masks.

Inferlet	What it shows
prefix-tree-kv-cache	Two-level copy-on-write branching from shared prefixes.
sliding-window-attention	Decode while attending only to the most recent token window.
attention-sink	Sink + sliding window (StreamingLLM).

Related guide

• Pages: the page model and committed-vs-working distinction.
• Forking and saving: copy-on-write branching and named snapshots.
• Inputs: BRLE attention masks for sliding windows and sinks.