orbitalo/homelab-brain
1
0
Fork 0
Code Issues 37 Pull requests Projects Releases Packages Wiki Activity Actions

refactor(llm): GPU-Architektur - Text lokal, Vision Cloud

Browse source 
- MODEL_VISION von qwen3-vl:32b (Ollama) auf openai/gpt-4o-mini (OpenRouter)
- Vision-Modell aus OLLAMA_MODELS entfernt (kein GPU-Swap mehr)
- keep_alive=-1 fuer Ollama: Textmodell bleibt permanent im VRAM
- RTX 3090 wird dedizierter Text+Embedding-Server
- Neue Dienste koennen Ollama mitnutzen ohne Konkurrenz
This commit is contained in:
Homelab Cursor 2026-03-25 20:56:55 +01:00
parent 5b57bfd27f
commit a0724ba6f1
1 changed files with 3 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
homelab-ai-bot/llm.py
View file

@ -19,11 +19,11 @@ OLLAMA_BASE = "http://100.84.255.83:11434"
OPENROUTER_BASE = "https://openrouter.ai/api/v1" OPENROUTER_BASE = "https://openrouter.ai/api/v1"
MODEL_LOCAL = "qwen3:30b-a3b" MODEL_LOCAL = "qwen3:30b-a3b"
MODEL_VISION = "qwen3-vl:32b" MODEL_VISION = "openai/gpt-4o-mini"
MODEL_ONLINE = "perplexity/sonar" MODEL_ONLINE = "perplexity/sonar"
FALLBACK_MODEL = "qwen2.5:14b" FALLBACK_MODEL = "qwen2.5:14b"
MAX_TOOL_ROUNDS = 3 MAX_TOOL_ROUNDS = 3
OLLAMA_MODELS = {MODEL_LOCAL, MODEL_VISION, FALLBACK_MODEL} OLLAMA_MODELS = {MODEL_LOCAL, FALLBACK_MODEL}
PASSTHROUGH_TOOLS = {"get_temperaturen", "get_energie", "get_heizung"} PASSTHROUGH_TOOLS = {"get_temperaturen", "get_energie", "get_heizung"}
@ -269,6 +269,7 @@ def _call_api(messages: list, api_key: str, use_tools: bool = True,
headers = {"Content-Type": "application/json"} headers = {"Content-Type": "application/json"}
timeout = _ollama_timeout_for(chosen) timeout = _ollama_timeout_for(chosen)
_add_no_think(payload.get("messages", [])) _add_no_think(payload.get("messages", []))
payload["keep_alive"] = -1
else: else:
url = f"{OPENROUTER_BASE}/chat/completions" url = f"{OPENROUTER_BASE}/chat/completions"
headers = {"Authorization": f"Bearer {api_key}"} headers = {"Authorization": f"Bearer {api_key}"}