Updating Llama Service Configuration | Generated by AI
Here’s a summary of what we did:
Service Setup
- Found 2 existing llama services:
llama-server.service(working) andllama.service(broken — wrong path with capitalP, kept crash-looping) - Updated
llama-server.serviceto match the actual running process settings instead of the outdated config
Config Change
| Before | After | |
|---|---|---|
| Model | DeepSeek-R1-0528-Qwen3-8B-Q4_K_M | gemma-4-E4B-it-Q5_K_M |
| Port | 11435 | 8081 |
| GPU layers | 99 | 43 |
| Context | 8192 | 65536 |
| Flash attn | — | off |
Things worth noting
llama.serviceis still enabled but broken — it’ll keep failing on every boot. Worth either fixing the path or disabling it:sudo systemctl disable llama.service- The service is still enabled, so it will auto-start on next reboot — remember to start it manually when you’re ready (or when the server is further away!)
- GPU dropped from ~5.7GB used → 166MiB after stopping, temp went idle at 35°C