Review or Reviews
테크, 개발, AI, 하드웨어 — 실사용 기반 리뷰와 가이드
최신 글
LLM·추론 최적화 문헌 모니터 — 2026-08-24
arXiv에서 수집한 로컬 LLM·추론 최적화 논문을 관련성 게이트로 거르고 로컬 LLM으로 구조화 요약한 2026-08-24 자동 피드입니다 (채택 36편 · 신규 16편). LLM 요약은 미검증 참고용이며, 인용 전 원문 확인이 필요합니다.
LLM·추론 최적화 문헌 모니터 — 2026-08-23
arXiv에서 수집한 로컬 LLM·추론 최적화 논문을 관련성 게이트로 거르고 로컬 LLM으로 구조화 요약한 2026-08-23 자동 피드입니다 (채택 36편 · 신규 7편). LLM 요약은 미검증 참고용이며, 인용 전 원문 확인이 필요합니다.
I Built a Hallucination Gate for My Own Writing. The Bigger Local Model Didn't Catch More
A 40-claim eval for a local hallucination checker on 8B, 9B and 27B. The 27B caught fewer than the 9B, the union of the two small models beat every single model, and a frontier judge finished one claim ahead.
더 보기
The Open-Model Cost Chart Everyone's Sharing Is API Prices. Here's What Self-Hosting Actually Gets You (Measured)
The intelligence-vs-cost chart making the rounds shows open models winning the value quadrant. True, but the x-axis is API token price. The cheap open winners are 100B-to-1T MoEs you can't run on a desktop GPU. Here's what you can actually self-host on an 11GB and a 24GB card, measured, and where the real ceiling is.
I Added a Verify Layer to My Local RAG to Catch Hallucinations. It Caught Me Being Wrong Twice About My Own Corpus
A claim-verification layer for a local RAG co-scientist, inspired by Karpathy's llm-wiki pattern. I tried to measure whether it catches hallucinations, almost shipped a false finding, and ended up with a clearer picture of what claim-checking can and can't do: it reliably catches values that are absent from the context, misses a real number pinned to the wrong question, and misses a false premise outright, and a model can't reliably referee its own blind spots.
What Actually Runs Well on a GTX 1080 Ti in 2026 (Measured)
The 'GPU poor' narrative says 24GB-and-below cards are eating well now thanks to QAT and MTP. But what about an 8-year-old 11GB GTX 1080 Ti? I measured it: Gemma 4 12B QAT at ~32 tok/s, Qwen3 8B at ~46, all fully on the GPU. Here's the table and where the ceiling is.
MTP Isn't Always a Win: 1.95× on My 3090, but Speculative Decoding Is Hardware-Dependent
MTP gave Gemma 4 12B QAT a 1.95x generation speedup on my 3090. But the same model with the same MTP draft runs 0.87x — slower — on an M1 Max. Speculative decoding is a hardware-dependent lever, not a free switch. Here are the measured numbers and why the draft-to-verify ratio decides it.
Gemma 4 QAT on a 1080 Ti: What 'Quantization-Aware' Actually Buys — and Fitting the 12B on 8 GB at 16k
QAT is the buzz around Gemma 4, so I ran it on actual old hardware. The quality claim holds up (vs naive Q4), the speed win is modest (~9%), and yes — you can run the 12B on an 8 GB card at 16k context. Here are the measured numbers and the exact recipe.
The Prefill Wall: Why MTP's 2× Barely Moves Long-Context Latency (Qwen3.6-27B, RTX 3090)
My last post doubled generation with MTP. A reader asked the question I'd skipped — what about prompt processing at long context? I measured prefill across context sizes on a 3090: a 64k prompt takes ~59s before the first token, and MTP can't touch that. Here's the math on when MTP's 2× actually matters, and when prefill swallows it.