<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Llm on History</title><link>https://kukie.tech/tags/llm/</link><description>Recent content in Llm on History</description><generator>Hugo</generator><language>ko-kr</language><lastBuildDate>Tue, 03 Nov 2026 10:00:00 +0900</lastBuildDate><atom:link href="https://kukie.tech/tags/llm/index.xml" rel="self" type="application/rss+xml"/><item><title>Kukie Serving.03 ServedModel Operator</title><link>https://kukie.tech/posts/kukie-serving-03/</link><pubDate>Tue, 03 Nov 2026 10:00:00 +0900</pubDate><guid>https://kukie.tech/posts/kukie-serving-03/</guid><description>Building kukie-oven, a Go operator that turns one ServedModel into a PVC, a download Job and a vLLM Deployment — the reconcile order, two server-side-apply field managers, telling someone else&amp;rsquo;s object from a leftover, Helm and GHCR packaging, an EBS CSI role limited by a tag boundary, and the first real run on an L40S: Ready in 14 minutes 11 seconds with no manual step.</description></item><item><title>Kukie Serving.02 GPU Add-on Design</title><link>https://kukie.tech/posts/kukie-serving-02/</link><pubDate>Tue, 03 Nov 2026 09:00:00 +0900</pubDate><guid>https://kukie.tech/posts/kukie-serving-02/</guid><description>Why an external LLM API plan became a Helm add-on for users&amp;rsquo; GPU clusters — quality and cost numbers behind the pivot, three layers split by required permissions, a chart with per-layer switches and a preflight job, the measurement-to-429 loop, an honest comparison with KubeAI and LiteLLM, and what is built versus still planned.</description></item><item><title>Kukie GitOps.04 Finding Helm Values</title><link>https://kukie.tech/posts/kukie-gitops-04/</link><pubDate>Tue, 03 Nov 2026 02:00:00 +0900</pubDate><guid>https://kukie.tech/posts/kukie-gitops-04/</guid><description>Finding which values key controls a rendered field in 21 public Helm charts — judging keys by marker rendering, narrowing candidates, why 18 of 209 fields were missed, comparing code with an LLM on five charts, and what a single render costs.</description></item><item><title>Kukie Serving.01 Baseline on One L40S</title><link>https://kukie.tech/posts/kukie-serving-01/</link><pubDate>Mon, 26 Oct 2026 02:00:00 +0900</pubDate><guid>https://kukie.tech/posts/kukie-serving-01/</guid><description>Measuring how many concurrent requests one NVIDIA L40S serves within a latency SLA before building a GPU cluster — the SLA draft as TTFT and inter-token latency, a single g6e.xlarge with two stop guards, a measurement script split from the server so it can run as a Kubernetes Job, results for four open models, why MoE models outrun a dense 27B on memory bandwidth, KV cache FP8 for 20% more headroom, and two review fixes.</description></item><item><title>Concurrency</title><link>https://kukie.tech/wiki/concurrency/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://kukie.tech/wiki/concurrency/</guid><description>같은 순간 LLM 서버가 처리 중인 요청 수. 늘리면 처리량은 늘다가 평평해지고 지연은 어느 지점부터 꺾여 오른다. SLA 를 넘기 직전 값이 GPU 한 장이 받는 양이다.</description></item><item><title>Goodput</title><link>https://kukie.tech/wiki/goodput/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://kukie.tech/wiki/goodput/</guid><description>SLA 를 지킨 요청만 센 처리량. 처리량은 끝낸 요청을 모두 세지만, goodput 은 늦은 요청을 실패로 보고 뺀다. 서빙 설정을 비교할 때 기준으로 쓴다.</description></item><item><title>ITL</title><link>https://kukie.tech/wiki/itl/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://kukie.tech/wiki/itl/</guid><description>LLM 이 토큰 하나를 내고 다음 토큰을 낼 때까지의 간격. 요청 하나의 평균 간격은 TPOT 이라 부른다. 사용자에게는 글자가 나오는 속도다.</description></item><item><title>KV Cache</title><link>https://kukie.tech/wiki/kv-cache/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://kukie.tech/wiki/kv-cache/</guid><description>LLM 이 이미 읽은 토큰의 Key·Value 를 GPU 메모리에 저장해 두는 캐시. 요청마다 토큰 수만큼 자라며, 남은 GPU 메모리가 동시에 받을 수 있는 요청 수를 정한다. PagedAttention 과 FP8 저장으로 자리를 아낀다.</description></item><item><title>MoE</title><link>https://kukie.tech/wiki/moe/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://kukie.tech/wiki/moe/</guid><description>Mixture of Experts. 트랜스포머의 FFN 을 여러 전문가로 나누고 라우터가 토큰마다 몇 개만 골라 계산하는 구조. 전체 파라미터는 크지만 토큰 하나에 쓰는 활성 파라미터가 작아, 메모리 대역폭에 묶이는 decode 가 빠르다.</description></item><item><title>TTFT</title><link>https://kukie.tech/wiki/ttft/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://kukie.tech/wiki/ttft/</guid><description>LLM 에 요청을 보낸 뒤 첫 토큰이 올 때까지의 시간. 대기 줄과 입력 처리(prefill)를 합친 값으로, 사용자에게는 글자가 나오기 시작하는 시간이다.</description></item></channel></rss>