Runbook — FASE 20 Router Qwen 2.5

Objetivo

Verificar que el AI-LAB Router usa los modelos correctos segun la ruta y que los wrappers legacy estan limpios en rutas no cognitivas.

Modelos por ruta

Ruta	Modelo esperado	max_tokens
minimal/casual	llama-3.1-8b-instruct	~96
minimal/greeting	llama-3.1-8b-instruct	~96
minimal/report	llama-3.1-8b-instruct	~180
observe	llama-3.1-8b-instruct	~180
fast	qwen/qwen2.5-coder-14b-instruct	256
general	qwen/qwen2.5-coder-14b-instruct	768
coding	qwen/qwen2.5-coder-14b-instruct	768
reasoning	qwen2.5-coder-32b-instruct	1200
tool_use	qwen/qwen3.6-27b	768

Checks rapidos

Router (8083)

# minimal
curl -sS -D- http://127.0.0.1:8083/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model":"auto","messages":[{"role":"user","content":"hola"}],"stream":false,"max_tokens":16,"temperature":0}' | grep x-ai-lab-selected-model

# fast
curl -sS -D- http://127.0.0.1:8083/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model":"fast","messages":[{"role":"user","content":"que es docker"}],"stream":false,"max_tokens":16,"temperature":0}' | grep x-ai-lab-selected-model

Gateway (8008)

# casual con tools
curl -sS http://127.0.0.1:8008/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model":"auto","tool_choice":"auto","tools":[{"type":"function","function":{"name":"question","description":"ask"}}],"messages":[{"role":"user","content":"podrias decirme que puedes hacer?"}],"stream":false,"max_tokens":32,"temperature":0}' | python3 -m json.tool | grep -E '"model"|prompt_tokens'

# general
curl -sS http://127.0.0.1:8008/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model":"auto","messages":[{"role":"user","content":"que es docker"}],"stream":false,"max_tokens":64,"temperature":0}' | python3 -m json.tool | grep -E '"model"|prompt_tokens'

Sintomas y soluciones

Sintoma	Causa probable	Accion
`qwen3.6-27b` en ruta general	`choose_model()` sin override	Verificar FASE 20B en gateway
502 con `no usable choices`	Router directo a LM Studio	Usar gateway (8008)
`prompt_tokens > 1000` en minimal	`max_tokens` override bug	Verificar FASE 20B en gateway
Respuesta con `[HARD_FACTS]` en fast	`build_system_context()` incorrecto	Verificar FASE 20B en router

Rollback

# FASE 20A
cp /opt/ai-lab/snapshots/fase20a-backup/* /opt/ai-lab/runtime/llm/
cp /opt/ai-lab/snapshots/fase20a-backup/inference_nodes.json /opt/ai-lab/config/

# FASE 20B
cp /opt/ai-lab/snapshots/fase20b-backup/router_api.py /opt/ai-lab/runtime/llm/
cp /opt/ai-lab/snapshots/fase20b-backup/openai_gateway.py /opt/ai-lab/runtime/gateway/

sudo systemctl restart ailab-router ailab-gateway

Panel de soporte

Dashboard: AI-LAB Route Family Observability Baseline
URL: http://192.168.1.40:3000/d/ai-lab-route-family/ai-lab-route-family-observability-baseline

Documentacion relacionada

/docs/historical/phases/fase-20a-migracion-qwen2.5-14b
/docs/fase-20b-limpieza-wrappers-legacy