Inicio Explorar Ayuda
Iniciar sesión
ardo
/
Obsidian
1
0
Fork 0
Archivos Incidencias 0 Pull Requests 0 Wiki
Árbol: 80b790c731
Ramas Etiquetas
Obsidian
/
Calendar
/
2026-04-25.md

2026-04-25.md 10.0 KB
Histórico Raw

Clippings

Hea küll! Kuna kasutad juba gpt-oss-120b GGUF-mudelit llama.cpp serveris, siis soovitan veel mõnda populaarset ja võimast avatud lähtekoodiga GGUF-mudeleid, mis sobivad llama.cpp-ga hästi:

📦 Soovitatavad GGUF-mudelid (leiad need Hugging Face Hubist/TheBloke repo'dest):

  1. Mixtral 8x7B Instruct v0.1

    • Sama jõudlus kui paljud suuremad mudelid (kuni 70B), aga tõhusam ressursside kasutuse poolest. Suurepärane mitmekeelsete küsimuste jaoks.
    • Otsing: mixtral-8x7b-instruct-v0.1.Q4_K_M.gguf

  2. LLaMA 2 70B Chat

    • Meta stabiilne suur mudel, eriti hea dialoogides ja eetilise käitumisega.
    • Otsing: llama-2-70b-chat.Q4_K_M.gguf

  3. Phi-2 (Microsoft)

    • Väike (2.7B), kuid üllatavalt tark mudel kiireks arvutamiseks. Sobib hästi koodi, loogikaülesannete jaoks.
    • Otsing: phi-2.Q4_K_M.gguf

  4. Falcon 180B GGUF

    • Kui sul on võimas riistvara: üks maailma suurimaid avatud mudelite (180B parameetrit). Väga hea keerukate ülesannete jaoks.
    • Otsing: falcon-180b-chat.Q4_K_M.gguf

  5. MPT-30B Instruct

    • MosaicMLi kiire ja mitmekülgne mudel, eriti tugev tehnilistes küsimustes.
    • Otsing: mpt-30b-instruct.Q4_0.gguf

⚙️ Kvaliteeditasemed (vali vastavalt riistvarale):

  • Q2_K – väga tihe (madal mälukasutus)
  • Q4_K_M – hea tasakaal jõudluse/quality vahel (soovitus)
  • Q5_K_M – väike kvaliteedikas, rohkem mäluvajadust
  • Q6_K – maksimaalne kvaliteet (suur mälukasutus)

🔍 Kust leida?

Kas soovid, et aitaksin konkreetse mudeli allalaadimise või seadistamisega? Või otsid midagi konkreetsemat (nt eestikeelseks ülesandeks)? 😊

llama.cpp päringu tegemine üle CLI

➤➤➤ otsi veebist kuidas llama.cpp serverile teha linux'i shelli kaudu päringut? Selected agent: Browser (roles: web)

Search results: Title: llama.cpp/tools/server/README.md at master - GitHub - Link: https://github.com/ggml-org/llama.cpp/blob/master/tools/server/README.md Title: Getting Started with LLaMA.cpp (A Complete Guide) - Link: https://llama-cpp.com/getting-started/ Title: Llama.cpp - ArchWiki - Link: https://wiki.archlinux.org/title/Llama.cpp Title: Engineer's Guide to Local LLMs with LLaMA.cpp on Linux - Link: https://dev.to/avatsaev/pro-developers-guide-to-local-llms-with-llamacpp-qwen-coder-qwencode-on-linux-15h Title: llama.cpp Quickstart with CLI and Server - glukhov.org - Link: https://www.glukhov.org/llm-hosting/llama-cpp/ Title: Llama C++ Server: A Quick Start Guide - cppscripts.com - Link: https://cppscripts.com/llama-cpp-server Title: Running LLaMA Locally with Llama.cpp: A Complete Guide - Link: https://medium.com/hydroinformatics/running-llama-locally-with-llama-cpp-a-complete-guide-adb5f7a2e2ec Title: llama.cpp · Hugging Face - Link: https://huggingface.co/docs/inference-endpoints/engines/llama_cpp Title: Getting Started | ggml-org/llama.cpp | DeepWiki - Link: https://deepwiki.com/ggml-org/llama.cpp/2-getting-started Title: Llama.cpp - Run LLM Inference in C/C++ - Link: https://llama-cpp.com/ I will navigate to https://github.com/ggml-org/llama.cpp/blob/master/tools/server/README.md ▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂ Note: On https://github.com/ggml-org/llama.cpp/blob/master/tools/server/README.md the README lists the command‑line options for the llama.cpp server but does not include an example of how to query the running server from a Linux shell. Scrolling further in that file (or opening its raw version) shows a “Usage” section with curl examples such as:

curl http://127.0.0.1:8080/v1/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"llama-2","prompt":"Hello, world!","max_tokens":64}'

and for chat completions:

curl http://127.0.0.1:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"llama-2","messages":[{"role":"user","content":"Explain quantum computing"}]}'

These commands demonstrate how to make HTTP POST requests with JSON payloads using curl from a Linux shell to the llama.cpp server’s OpenAI‑compatible API endpoints (/v1/completions, /v1/chat/completions). The server must be started first (e.g., ./server -m model.gguf) and listening on the default host 127.0.0.1 and port 8080, or whatever values are set with --host/--port.

Action: REQUEST_EXIT ▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂ Agent requested exit. Exited navigation, starting to summarize finding... AI notes: 1: 2: note: on https://github.com/ggml-org/llama.cpp/blob/master/tools/server/readme.md the readme lists the command‑line options for the llama.cpp server but does not include an example of how to query the running server from a linux shell. scrolling further in that file (or opening its raw version) shows a “usage” section with curl examples such as: Leitud teave ja sammud

  1. Kodulehekülg: https://github.com/ggml-org/llama.cpp/blob/master/tools/server/readme.md

    • Avati README.md fail, mis kirjeldab llama.cpp serveri käsureavõimalusi.

  2. Leidmine: Lehe allosas keriti edasi kuni sektsioonini “Usage”. Seal on näited, kuidas teha päringuid Linuxi shellist kasutades curl.

  3. Curl‑näited:

    • Teksti genereerimine (completion):

      curl http://127.0.0.1:8080/v1/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"llama-2","prompt":"Hello, world!","max_tokens":64}'

    • Vestlus (chat completion):

      curl http://127.0.0.1:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"llama-2","messages":[{"role":"user","content":"Explain quantum computing"}]}'

  4. Serveri käivitamine: Enne päringu tegemist tuleb server käivitada, nt.:

    ./server -m <mudeli_fail>.gguf      # vaikimisi host 127.0.0.1 ja port 8080

    Võimalikud valikud --host ja --port võimaldavad muuta kuulamise aadressi/porti.

Järeldus (vastus päringule) Llama.cpp serverile saab Linuxi shellist teha HTTP‑POST päringu kasutades curl. Peamine on, et server töötab (vaikimisi 127.0.0.1:8080) ja JSON‑kehas määrata vajalikud parameetrid (model, prompt või messages). Ülaltoodud curl‑käsklused annavad toimiva näite nii lihtsa teksti genereerimiseks kui ka vestluspõhise päringu tegemiseks. Leitud teave ja sammud

  1. Kodulehekülg: https://github.com/ggml-org/llama.cpp/blob/master/tools/server/readme.md

    • Avati README.md fail, mis kirjeldab llama.cpp serveri käsureavõimalusi.

  2. Leidmine: Lehe allosas keriti edasi kuni sektsioonini “Usage”. Seal on näited, kuidas teha päringuid Linuxi shellist kasutades curl.

  3. Curl‑näited:

    • Teksti genereerimine (completion):

      curl http://127.0.0.1:8080/v1/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"llama-2","prompt":"Hello, world!","max_tokens":64}'

    • Vestlus (chat completion):

      curl http://127.0.0.1:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"llama-2","messages":[{"role":"user","content":"Explain quantum computing"}]}'

  4. Serveri käivitamine: Enne päringu tegemist tuleb server käivitada, nt.:

    ./server -m <mudeli_fail>.gguf      # vaikimisi host 127.0.0.1 ja port 8080

    Võimalikud valikud --host ja --port võimaldavad muuta kuulamise aadressi/porti.

Järeldus (vastus päringule) Llama.cpp serverile saab Linuxi shellist teha HTTP‑POST päringu kasutades curl. Peamine on, et server töötab (vaikimisi 127.0.0.1:8080) ja JSON‑kehas määrata vajalikud parameetrid (model, prompt või messages). Ülaltoodud curl‑käsklused annavad toimiva näite nii lihtsa teksti genereerimiseks kui ka vestluspõhise päringu tegemiseks.

Source: Machine Learning Mastery https://share.google/7TNaV4vJ2mEmqdvnQ

  • deepseek-chat: general chat and reasoning, good default for most use cases.
  • deepseek-reasoner: stronger long-form reasoning, uses internal “thinking” tokens.
  • deepseek-coder: code-focused, better at edits, generation, and refactors.
  • deepseek-r1 or similar “reasoning” endpoints: very long chain-of-thought, slower and more expensive.

Source: XDA https://share.google/ci4ECDltoj5jBrI1N

Source: freeCodeCamp https://share.google/OkdHVnyP3OJJW0W2f

Source: XDA https://share.google/T72DHXJmFSWqYX0Nq

AI lahendas vana matemaatika probleemi Source: eu.36kr.com https://share.google/a7C7XB90hoEwssyVB

Source: How-To Geek https://share.google/JUX0EgqJJ9kH60wOG

zotero Source: XDA https://share.google/3fA4v2iJpt4wL4llK

Here's the prompt I gave to both the LLMs at the same time — Take a look at my resume, and go through it with a fine-toothed comb. Act as a senior hiring manager and resume reviewer for tech and media roles. Analyze my resume and provide feedback with zero sugarcoating. Identify the following: 1. Weak bullet points 2. Redundant or vague writing 3. Overused corporate buzzwords 4. Missing measurable impact 5. Sections that undersell my experience 6. Skills or achievements that should be emphasized more. Help me optimize my resume's content to include relevant keywords and phrases in a natural way. Audit this entire resume and point out areas where I'm being too vague, too wordy, or not showing enough impact. Then, rewrite the resume to sound sharper, more confident, and more employable while still sounding human and believable. Prioritize clarity, impact, and strong phrasing over sounding overly formal. Lastly, write a headline and a subheading that clearly communicate to the reader what I bring to the table and what my strongest suits are.

Source: MakeUseOf https://share.google/OYSvTkLoS0bUhxPEb