LiteLLM
Allikas: Imre kasutab arvutit
Mine navigeerimisribaleMine otsikasti
Sissejuhatus
TODO
Tööpõhimõte
TODO
Võib olla võiks AI teemat kujutada sellistes dimensioonides
- corporate nähtus mudeli taga (Anthropic, Google jt)
- mudel (sonnet, gemini jt)
- chat/app liides (https://claude.ai/, https://gemini.google.com/ jt)
- agent (Claude Code, Gemini CLI jt)
- api liides
kus
- TODO
Paigaldamine ja seadistamine
Paigaldamiseks sobib nt selline docker compose
TODO
Tulemusena käivitatakse selline komplekt konteinereid
root@dh-ubuntu-2604-litellm-01:/srv/litellm/dc# docker compose up -d [+] up 85/85 ✔ Image redis:7-alpine Pulled 31.8s ✔ Image ghcr.io/berriai/litellm:main-latest Pulled 49.6s ✔ Image nginx:alpine Pulled 33.3s ✔ Image postgres:16-alpine Pulled 36.6s ✔ Image ghcr.io/open-webui/open-webui:main Pulled 69.0s ✔ Network genai-internal-network Created 0.2s ✔ Volume p_litellm_redis_data Created 0.1s ✔ Volume p_litellm_litellm_data Created 0.1s ✔ Volume p_litellm_openwebui_data Created 0.1s ✔ Volume p_litellm_postgres_data Created 0.1s ✔ Container genai-postgres Healthy 13.1s ✔ Container genai-redis Healthy 12.6s ✔ Container genai-litellm-proxy Started 11.8s ✔ Container genai-openwebui Started 11.7s ✔ Container genai-nginx-ingress Started 11.1s root@dh-ubuntu-2604-litellm-01:/srv/litellm/dc#
kus
- TODO
LiteLLM
LiteLLM seadistamiseks veebipõhiselt
kus
- TODO
Open WebUI
Admin liides
Kasutajale mõeldud liides
llm-client - llm-server
llm-client ja llm-server vahelise suhtlemise jälgimine
root@dh-ubuntu-2604-litellm-01:~# tcpdump -s 0 -w ccd-20260727-03.pcap -ni ens18 port 4000
kus
- -s 0 - et võrgus liikuvad paketid täies mahus ära salvestuksid
ning edasi wireshark peal paistab
imreoolberg@Imres-MacBook-Air ~ % jq '.tools[].name' ccd-20260727-03-tmp.json | nl
1 "Agent"
2 "AskUserQuestion"
3 "Bash"
4 "CronCreate"
5 "CronDelete"
6 "CronList"
7 "DesignSync"
8 "Edit"
9 "EnterPlanMode"
10 "EnterWorktree"
11 "ExitPlanMode"
12 "ExitWorktree"
13 "Monitor"
14 "NotebookEdit"
15 "PushNotification"
16 "Read"
17 "ReportFindings"
18 "ScheduleWakeup"
19 "SendMessage"
20 "Skill"
21 "TaskCreate"
22 "TaskGet"
23 "TaskList"
24 "TaskOutput"
25 "TaskStop"
26 "TaskUpdate"
27 "WebFetch"
28 "WebSearch"
29 "Workflow"
30 "Write"
ning
imreoolberg@Imres-MacBook-Air ~ % jq '.tools[] | {name, description: (.description[0:80])}' ccd-20260727-03-tmp.json | head -n 12
{
"name": "Agent",
"description": "Launch a new agent to handle complex, multi-step tasks. Each agent type has spec"
}
{
"name": "AskUserQuestion",
"description": "Use this tool only when you are blocked on a decision that is genuinely the user"
}
{
"name": "Bash",
"description": "Executes a bash command and returns its output.\n\n- Working directory persists be"
}
Kasulikud lisamaterjalid
- TODO