Engineering notes
Articles for people who
ship the stack
Original rewrites on React, Node.js, TypeScript and AI — practical notes from the same engineering practice behind our operator software. Article bodies are in English.
Tagged: prefill
QUOKA Explained: Choosing Which KV History Attention Should Read
How QUOKA speeds up long-prompt prefill: KV cache basics, chunked prefill, picking outlier queries, max scoring, and cheap pre-aggregation for grouped-query attention.
2028 wordsRead articleMoving KV Cache Between Qwen3 Models Instead of Re-Prefilling
NVIDIA research maps Qwen3-14B attention state into Qwen3-32B with a ridge-regression bridge. See what the latency numbers mean, where it fails, and why agents care.
1817 wordsRead article
FAQ
About these articles
Yes. Each piece is an original English rewrite prepared for this site. Structure and code examples follow the source material; prose is rewritten and checked for overlap before export.
Technical posts ship in English first so code and API names stay exact. The site chrome (navigation, SEO shells) is available in all nine languages.
New drafts land in the publish queue as they pass editorial checks. Running the site import picks up only new or changed slugs — nothing is duplicated.
Treat samples as illustrative. Validate against your stack and licences before production use. For delivery work, contact us with the brief.
Request a 24h estimate
Need the same stack in a production operator layer? Send the brief — estimate within 24 hours.