koreallmdev/qwen3-8b-korean-coding-router

Qwen3 8B Korean Coding Router - Combined Local + Router Runtime

한국어 설명

Qwen3-8B를 Ollama에서 불러오고, 우리가 만든 Router API v4.1과 Open WebUI Docker stack을 결합한 한국어 코딩/운영 보조 런타임 배포판입니다.

English Summary

Qwen3-8B based Korean coding/ops runtime package for Ollama, Open WebUI, and Docker Router API v4.1. Includes local Ollama runtime and router stack; model weights are not included.

Stable Release

STABLE_LOCKED_VERSION=v4.1
ROUTER_API=Qwen3RouterAPI/4.1
ROUTING=priority_intent_router_v4_1
HF_REPO=koreallmdev/qwen3-8b-korean-coding-router

This repository combines two local runtime layers:

1. local_ollama_runtime/
   Local Ollama package with Modelfile, terminal wrapper, smoke scripts, policies, and reports.

2. docker_openwebui_router_stack/
   Docker Open WebUI + Ollama + Router API stack with the v4.1 priority intent router.

This repository is not a full Hugging Face model checkpoint and does not include Qwen3 base weights.

Stable version:

v4.1 combined local + router runtime
snapshot=20260711_234506

Required base model

ollama pull qwen3:8b

Option A: Docker Open WebUI + Router API

cd docker_openwebui_router_stack
bash start_stack.sh
bash create_ollama_model.sh
bash smoke_stack.sh

Open WebUI:

http://127.0.0.1:3001

Recommended model in Open WebUI:

qwen3-8b-korean-coding-router

Router API:

http://127.0.0.1:8011/v1

Ollama API exposed to host:

http://127.0.0.1:11435

Internal Docker addresses:

Open WebUI -> Ollama = http://ollama:11434
Open WebUI -> Router API = http://router-api:8011/v1
Router API -> Ollama = http://ollama:11434

The local OpenAI-compatible API key used by the Docker stack is a non-secret placeholder:

local

Option B: Local Ollama terminal/runtime package

cd local_ollama_runtime
bash create_ollama_model.sh
bash run_ollama_router_chat.sh

Static Ollama model:

qwen3-8b-korean-coding-router:latest

Quick run from repo root

Docker stack:

bash run_docker_router_stack.sh

Local Ollama terminal runtime:

bash run_local_ollama_runtime.sh

v4.1 router behavior

priority_intent_router_v4_1
route/debug endpoint
port/PID safety guard
Docker/Open WebUI stack URL guard
longform-context misroute prevention
history trimming for long answers

For port/PID questions, the router enforces:

확인 전에는 kill하지 않습니다.
ss -lntp
ps
kill -9는 마지막 수단입니다.

For current Docker/Open WebUI stack questions, the router prioritizes:

http://ollama:11434
http://router-api:8011/v1
http://127.0.0.1:3001

and avoids old host bridge answers such as:

host.docker.internal:11434
172.17.0.1:11434

Important notice

Do not load this repository with:

AutoModelForCausalLM.from_pretrained("koreallmdev/qwen3-8b-korean-coding-router")

This repository is a runtime package, not a model checkpoint.