베스트
-
15추천폐쇄망용 완전 오프라인 RAG. Ollama만으로 동작, 임베딩은 bge-m3. 피드백 환영합니다.
-
14추천
-
13추천
-
12추천가격 대비 성능이 미쳤다. $0.14/M in. 로컬 vLLM 대신 API로 갈아탈까 고민 중.
-
12추천링크2.8T 파라미터면 DeepSeek V4 Pro보다 75% 크다고 합니다. 1M 컨텍스트 + 항상 켜진 추론 모드. 오픈소스가 이제 안 따라잡히네요.
-
12추천이미지 링크8B~27B 모델을 서빙하려고 합니다. throughput이 중요하면 vLLM, VRAM이 빠듯하면 llama.cpp라는 얘기가 있는데 실제 운영하시는 분들 경험 공유 부탁드려요.
-
12추천
-
11추천L4 한 장으로 Qwen2.5-7B 서빙. 처리량 만족스럽습니다. 양자화 없이도 충분하네요.
-
11추천
-
11추천링크코딩 테스트 통과율이 예전보다 확실히 높아진 느낌. AI 도구 사용을 허용할지, 아니면 역량 평가 방식을 바꿔야 할지 고민입니다.
-
11추천
-
10추천LangGraph, CrewAI, 자체구현... 다들 뭐 쓰시는지 궁금합니다.
-
10추천
-
10추천링크LangGraph 같은 프레임워크 쓰다가 직접 while 루프로 짜는 게 낫다는 글을 봤는데요. 디버깅이 쉽고 의존성이 적다는 게 이유라네요. 다들 어떻게 생각하세요?
-
10추천이미지 링크로컬 RAG 서비스 최소 구성 템플릿을 만들어봤습니다. docker compose 하나로 Ollama + FastAPI + ChromaDB가 뜹니다. 피드백 환영해요.
-
9추천Ollama on a laptop is great for tinkering, but once you need reliability, hosted APIs win on latency and uptime. My experience so far.
-
9추천DeepSeek 캐시 히트율 66% 나오는데, 프롬프트 구조 어떻게 짜면 캐시가 잘 맞나요?
-
9추천MathBench leaderboard flipped again. The open-source gap is closing faster than expected.
-
9추천웹에서 문서를 읽어와서 요약하는 기능인데, 문서에 악성 지시문이 있으면 대응이 어렵네요. 방어 전략 공유 부탁드려요.
-
9추천보안 때문에 전부 온프레미스로. bge-m3 + Qwen2.5-14B로 구축했고, 사용자 만족도 4.2/5 나왔어요. 후기 공유합니다.
-
9추천Ticket deflection went from 0% to 38% in a month. Here's the architecture and the mistakes we made.
-
9추천링크솔직히 이번 건 좀 궁금하네요. 정부가 어떻게 안전하다고 결론 내렸는지 과정이 투명하지 않다는 지적. 원문 공유합니다.
-
9추천
-
9추천
-
9추천
-
9추천이미지 링크인터넷이 안 되는 환경에서 LLM+RAG를 돌려야 하는 프로젝트였는데, Ollama + 로컬 임베딩으로 구성했습니다. 오프라인 배포에 관심 있으신 분들께 도움이 되길.
-
9추천이미지 링크They released the evals and methodology this time, which is rare. HumanEval pass@1 is a couple points ahead of Llama 4.5 and it's faster to serve. We're spinning up a test pod this week. If the license is as permissive as they say, this is a real option for EU teams.
-
8추천한국어 문서 검색인데 bge-m3 vs multilingual-e5 고민입니다. 경험 공유 부탁해요.
-
8추천로컬에서 돌릴 수 있는 새 오픈웨이트 모델 소식. 7B인데 성능이 무섭다.
-
8추천드디어 에이전트 전용 API가 나왔네요. MCP 서버가 기본 지원이라 도구 연동이 쉬워질 듯. 개발자 문서부터 읽어봐야겠어요.
-
8추천도메인 용어가 많은데 파인튜닝이 나을지 RAG가 나을지. 둘 다 해보신 분들 의견 궁금합니다.
-
8추천Every framework becomes a debugging nightmare past 500 lines. Plain code + function calling has been better for us.
-
8추천비용/지연/프롬프트 버전까지 한 번에 보이니까 디버깅 시간이 반으로 줄었습니다. 무료 티어로 충분해요.
-
8추천듀얼 3090으로 70B 돌리려다 이틀 삽질. 결국은 OLLAMA_NUM_GPU + 스플릿 모드 조합으로 해결. 정리해봤습니다.
-
8추천금융권 고객사에 설치했습니다. 인터넷 완전 차단 환경에서도 임베딩+검색+생성 전부 동작. 자세한 내용은 글에.
-
8추천Replaced a custom fine-tune with a generic 8B + good retrieval. Cheaper to maintain, better scores. Data point for you.
-
8추천링크Mira Murati가 세운 회사가 시드로 2조원대 유치. a16z 리드. 시장 분위기가 확실히 달라졌네요.
-
8추천이미지 링크코드 문서는 300토큰, 일반 문서는 500~800토큰으로 쓰고 있는데요. 법률 문서처럼 구조화된 건 섹션 단위로 자르는 게 낫다는 의견도 있더라고요. 다들 어떻게 하세요?
-
8추천링크사내용 챗봇 하나 돌리는데 GPU 서버 빌리는 게 월 200만원이 넘네요. 트래픽이 적으면 API 쓰는 게 이득이라는 의견도 있고요. 실제 사례 있으면 공유 부탁드려요.
-
8추천
-
8추천
-
8추천이미지 링크Woke up to the leaderboard shuffle. Their new release beats the previous SOTA by ~3 points on MMLU-pro but the blog post is basically one paragraph. Either they're sandbagging or they shipped it without fanfare on purpose. Either way the pricing page is still the old one, which is the real story.
-
8추천이미지 링크Weights are up on the hub. The 8B with the new tokenizer is suspiciously good at tool calling for its size. Fine-tuning on our internal data took 6 hours on a single H100. If you need a cheap default for agents, this is hard to beat.
-
8추천Our org has 10k pages of policy PDFs nobody reads. Built a RAG bot with citations that links back to the exact page. Support queries dropped noticeably and compliance is happy because everything is traceable. Side project that became the team's favorite tool.
-
7추천Feed it a CSV export, it finds anomalies and writes a Korean explanation via an LLM. Useful for MSP folks.
-
7추천가성비 미쳤습니다. 4bit로 12GB 램 노트북에서 돌아간다고 하니 로컬 실험용으로 딱이네요.
-
7추천Prefill pipelining finally landed. Our 128k-context serving costs should drop meaningfully.
-
7추천현재 Chroma 쓰다가 Qdrant로 옮길지 고민 중. 데이터 500만 건 정도인데 경험 있으신 분?
-
7추천bge-m3에서 다른 모델로 바꾸려는데, 기존 문서 전부 다시 인덱싱해야 하는 게 맞는지요?
-
7추천Mine: Claude for code, DeepSeek for bulk tasks, local Ollama for privacy stuff. What's yours?
-
7추천일 10만 요청 기준 DeepSeek API vs 자체 GPU. 대략 어디서 크로스오버되나요? 계산해보신 분?
-
7추천네임스페이스 API가 안정화됐고 성능도 올라갔다고 합니다. 마이그레이션 가이드 공유합니다.
-
7추천Ran a quick benchmark. pgvector with HNSW was 80% the speed at 10% the cost. Details in thread.
-
7추천All data stays on device. Weekly digest with topics, mood, and action items. 400 users in 2 weeks!
-
7추천동일 7B 모델로 비교했는데 p50은 비슷하고 p99는 TRT-LLM이 더 안정적이었습니다. 수치 표로 정리했어요.
-
7추천링크Multiple model sizes, powering products across Meta. The open-vs-closed debate keeps heating up.
-
7추천링크Updated list of open LLMs you can use commercially. Handy when picking a model for a client project.
-
7추천
-
7추천
-
7추천이미지 링크둘 다 써봤는데, Open-webui는 Ollama 연동이 편하고 LibreChat는 멀티모델 관리가 좋았습니다. RAG 기능은 LibreChat가 좀 더 성숙한 느낌? 여러분 선택은?
-
7추천
-
7추천이미지 링크Downloaded the 27B variant last night. Quantized to Q4 it fits in 16GB RAM and generates at a very usable speed. Not Claude-level reasoning, but for a local model it's a big step. Google is winning the 'open weights you can actually run' game right now.
-
7추천이미지 링크Read the paper twice. The training cost figures are almost insulting compared to US labs. Latency is great, price is lower than anyone expected. Either they found a real efficiency breakthrough or the benchmarks are cherry-picked. I'd love to see third-party replications.
-
7추천Day 1: I felt like a god. Day 12: I found a bug the AI introduced that took 3 hours to hunt. Day 30: net positive, probably. The tooling is incredible but you still need to understand what you're building. No replacement, just acceleration.
-
7추천링크50 lines of TypeScript, points at our search API, and now everyone queries docs from their editor. It's dumb but it changed how we work. The protocol is clunky but the integration win is real.
-
7추천Collected 3 years of my blog posts and newsletter, fine-tuned a 7B with LoRA. The style imitation is uncanny — my wife thought I wrote the drafts. Uses: first drafts of posts I'm too lazy to start. It's my ghostwriter now.
-
7추천12B model on the family PC, a simple web UI. It explains math problems step by step instead of just giving answers. No accounts, no data leaving the house, and no monthly subscription. The kids actually use it.
-
6추천Chroma vs Qdrant vs LanceDB for a small self-hosted RAG service. Disk and RAM matter — running on a free-tier VM.
-
6추천Scrapes arXiv + HN, summarizes with an LLM, posts as threads. 200 stars in a week. AMA.
-
6추천국내+글로벌 AI 논의를 한 곳에서. 짧게 쓰레드처럼 쓰는 곳입니다. 자유롭게 이야기해요.
-
6추천Big claim. If real, this changes agent UX a lot. Anyone benchmarked it yet?
-
6추천오프라인 어시스턴트가 실생활에 들어오는 시점이네요. 개인정보 처리 관점에서도 의미 있는 변화.
-
6추천LLM이 스트리밍으로 출력할 때 JSON을 점진적으로 파싱하는 방법 질문입니다. partial JSON 파서 쓰시는 분 계신가요?
-
6추천Long agent sessions blow up tokens. Do you use summarization, sliding windows, or something else?
-
6추천Model cites chunks that don't actually support the answer. Any post-processing tricks that work?
-
6추천LLM을 잘 쓰는 것만으로는 부족하다는 말이 많아지네요. 시스템 설계? 벡터 DB? MLOps? 다들 뭐가 우선이라 생각하세요?
-
6추천보안팀이랑 임원분들이 '아직 이르다'는 입장인데, 실무는 이미 AI 없이는 일이 안 됩니다. 설득 팁 있을까요?
-
6추천With better rerankers and hybrid search, raw embedding quality matters less every quarter. Agree?
-
6추천We went LibreChat for multi-user + RBAC. Open-webui has nicer UX though. Your pick?
-
6추천드래그앤드롭으로 LLM + 검색 + 슬랙 연동까지. 개발자도 가끔은 로우코드가 편하네요.
-
6추천CSV of prompts → JSONL of responses + scores. 100 lines of Go. Maybe useful for your eval pipeline.
-
6추천Scrapes 20 sources, dedupes, summarizes, posts at 8am. 1.2k subscribers now. AMA about the pipeline.
-
6추천주 1회 LLM 논문 리뷰. 누적 12만 뷰 달성. 글쓰기 + 실험 재현을 병행하니 배우는 게 두 배네요.
-
6추천링크Doctors' AI scribe just raised a huge round at $5.3B valuation. Healthcare AI is the real money-maker apparently.
-
6추천
-
6추천
-
6추천링크보안 우려로 LLM 도입을 꺼리시는 분들이 있는데, 폐쇄망 배포 사례를 보여드리면 좀 나아질까요? 비슷한 경험 있으신 분들의 조언 부탁드립니다.
-
6추천
-
6추천이미지 링크Just got the email. Cache writes are 50% cheaper and cache reads dropped too. For anyone running long agent loops this is basically free money. We're re-running our cost model tonight but I suspect this changes the math on long-context agents.
-
6추천이미지 링크Ordered 8 cards for our inference cluster two weeks ago. Sales rep just told me new orders are looking at Q4. Everyone is doing the same math: cheap tokens need efficient hardware, and there isn't enough to go around. If you're planning capacity for next year, order now.
-
6추천Ship a model, top the leaderboard, get the press release, repeat. Meanwhile real usage is full of edge cases none of the benchmarks touch. I'd rather have a model that's honest about what it can't do.
-
6추천Two months in, no regrets so far. Working on an open agent framework and living off savings. The open source ecosystem is chaotic but the pace is addictive. Happy to answer questions about the leap, money, or what I actually do all day.
-
6추천링크We were merging PRs blind. Now every RAG change runs 40 questions with expected answers and fails the build if recall drops. It's crude but it caught two regressions last week. Sometimes the simple stuff wins.
-
6추천링크Ran the same 1M-doc dataset through all three. Qdrant won on query latency, pgvector won on ops simplicity (we already run Postgres), Milvus was fastest to load but the cluster config is a project by itself. Full numbers in the post.
-
6추천Workshop is loud, dusty, and has no wifi worth trusting. Pi 5 + a local model + a cheap mic array. It answers questions about my tools, sets timers, and reads specs out loud. Everything runs on-device and it cost under $200.
-
6추천링크Been dumping notes, highlights, and bookmarks into a local RAG setup for two years. Search is instant and surprisingly good. Open-sourced the whole thing so it's reproducible. Warning: your old notes are more embarrassing than you remember.
-
6추천Scrapes my git commits and PRs, summarizes them into 3 bullet points, emails them before standup. Takes 30 seconds of my morning. My manager thinks I'm extremely organized. I'm just lazy with extra steps.
-
6추천It joins the call with a normal name, takes notes, and posts them to our wiki. People forget it's there until the notes appear. Uses a tiny local model for transcription. The 'is this legal' question came up once. HR said fine, so we're shipping it.
-
5추천I keep coming back to plain code + function calling. Frameworks add too much magic for debugging.
-
5추천RAG 기반 기업 검색 솔루션으로 유치했다고 합니다. 국내도 이제 본격적으로 크네요.
-
5추천Small enough to self-host, strong enough for real agent loops. Worth a look for cost-sensitive teams.
-
5추천We use RAGAS but the scores don't correlate with human judgment. What do you actually use in production?
-
5추천vLLM vs TGI vs llama.cpp server. For a small team serving one 8B model, what's the least ops burden?
-
5추천임베딩 + 소형 LLM만으로 API 없이 운영 가능한지 궁금합니다. 트래픽은 하루 1천 건 수준이에요.
-
5추천코딩 문제는 다들 잘 풀어요. 문제는 '왜'를 물었을 때. 면접 방식 자체를 바꿔야 하나 고민입니다.
-
5추천I used to read every arxiv paper. Now I wait for good summaries. Am I falling behind?
-
5추천새 모델 나올 때마다 다같이 벤치마크 까보는 게 제일 재밌네요. 여기 오시는 분들은 뭐 보러 오세요?
-
5추천I packaged my go-to RAG setup into a template. Docker compose up and you have a working API in 2 minutes.
-
5추천지금은 Git으로 관리하는데, 버전별 A/B 테스트가 필요해졌습니다. 전용 도구 추천 부탁해요.
-
5추천Gradio got multimodal chat components. Streamlit is catching up. For internal demos, which wins?
-
5추천Visual node editor for agent pipelines with live tracing. MIT licensed, would love feedback.
-
5추천매주 토요일 2시간. 30명에서 시작해 지금 12명의 코어 멤버로. 꾸준함의 가치를 배웠습니다.
-
5추천링크34 of the 100 fastest-growing companies are AI. The capital shift is real and it's accelerating.
-
5추천링크Both raised big rounds in 2025. If you're building on top of LLMs, infra plays look safer than apps.
-
5추천링크Keep this tab open. Everything from o1 to GPT-5-Codex-Mini lands here first.
-
5추천
-
5추천이미지 링크The grace period is ending. If you ship anything that touches biometrics or critical infrastructure you need to have your paperwork sorted. Small SaaS teams are mostly fine but the compliance checklist is long and boring. Start with the Annex III list, don't assume you're exempt.
-
5추천이미지 링크The 3B model handles summarization and smart replies locally, and it's snappy. Privacy story is nice. Devs can call it via a new CoreML API. It's not going to replace cloud models for hard tasks, but for simple stuff it's genuinely useful and free.
-
5추천Building evals feels like a second product. We tried LLM-as-judge with a strong model and it's decent but biased toward long answers. Anyone using lightweight checks + sampling? What's your minimal viable eval?
-
5추천We're about to onboard our first enterprise customers and I have to decide. Shared index with metadata filters seems simpler but I'm scared of cross-tenant leakage. Per-tenant collections mean more ops. What do you run?
-
5추천The agent handles maybe 60% of tickets but the 40% it hands to humans are the messy ones, so support load barely dropped. Customers who realize they're talking to a bot get annoyed even when it works. Not sure this was a win.
-
5추천We priced our product assuming $2/M input tokens. Now we're at $0.30 and dropping. Great for us, but customers keep asking why their bill didn't go down if AI got cheaper. Also, our competitors can now afford the same quality. Margin moat is shrinking.
-
5추천링크LangSmith is smoother but the per-seat pricing hurt us. Langfuse self-hosted costs us one small VM. Missing a few features but the trace UI is good enough and the data stays on our infra. For EU clients that matters.
-
5추천링크Wrote our own gateway instead of buying one. Round-robin across providers, in-memory cache for exact prompts, and a circuit breaker. It's not fancy but it cut our API bill 35% and we control everything.
-
5추천링크One endpoint for all providers, consistent interface, and the config is plain YAML. We swapped out three vendor SDKs in an afternoon. If you're multi-provider, just use it.
-
5추천Like this site but for semiconductor news specifically. Scrapes 50 RSS feeds, dedupes with embeddings, summarizes with an LLM, ranks by my reading habits. It's been my morning read for a month and I've never opened the old feed reader again.
-
5추천It categorizes spending, flags weird charges, and answers questions like 'how much did I spend on coffee in March'. Accuracy is 90%+ on categories. Caveat: you have to trust it with your data, so it runs fully local. Worth it for the anxiety reduction alone.