TGViewer
Aziz Mirzayev Aziz Mirzayev @pro_aim · 938 subscribers
Post #1738 265
Ollama vs vLLM vs SGLang. Uchchalasi ham open model'larni ishga tushiradi, lekin har biri boshqa ish uchun:

— Ollama: bitta user, bitta mashina. Laptop/Mac'da prototip qilish uchun. Anthropic API formatini ham tushunadi, ya'ni Claude Code'ni local model bilan ishlatsa bo'ladi
— vLLM: ko'p user, har xil prompt'lar. Continuous batching + PagedAttention
— SGLang: agent'lar, RAG, uzun chat'lar. Umumiy prefix'larni qayta ishlatadi (RadixAttention)

Bonus tip: system prompt va tool'larni har safar bir xil qoldiring, o'zgaradigan qismni oxiriga qo'ying. Shunda prefix cache ishlaydi.

Sizda local model'lar qaysi birida ishlaydi?

Manba: https://lnkd.in/p/dfJYjuK9
LinkedIn Ollama vs vLLM vs SGLang: Choosing the Right Model for Your Use Case | Science posted on the topic | LinkedIn Ollama vs vLLM vs SGLang They all run open models. Each one is built for a different kind of traffic. ⚙️ Ollama: one user, one machine → Requests wait in a queue, 1 per model at a time by default → Runs quantized models: GGUF via llama.cpp, and MLX by default…
  • 👍 2
More from @pro_aim
  1. Oct 11, 2026Harness nima ekanini 24 daqiqada tushuntirib beradigan yaxshi video ekan. Claude Code va C…
  2. Oct 11, 2026RAG noto'g'ri javob bersa, ko'pincha LLM emas, retrieval aybdor ekan. O'xshash chunk degan…
  3. Oct 11, 2026AI agent'lar uchun 20 ta open-source security tool. LinkedIn'da shunaqa ro'yxat chiqibdi �…
  4. Oct 11, 2026🇺🇿 UEnter Higgsfield AI yordamida video yaratish bo‘yicha kursga qabul ochmoqda Sun’iy i…
  5. Oct 11, 2026🇺🇿 Buxoroda Plug and Play’ning xalqaro inkubatsiya dasturiga arizalar qabuli ochildi G‘o…
  6. Oct 11, 2026🏆 O‘zbekiston — robototexnika bo‘yicha jahon chempioni! 🇰🇷Janubiy Koreyaning Incheon sh…
Threads Profile ViewerView any public Threads profile without an account.Open ThreadLook →Writing with AI? Make it sound human.Metric37 rewrites AI drafts so they read naturally. Free AI detector, 1,500 words free.Try Metric37 →