healthCheckTimeout: 60 models: kokoro-tts: proxy: http://kokoro-tts-server.llm.svc:8880 checkEndpoint: /v1/audio/speech unlisted: true gemma-3-12b-it-qat: cmd: > /app/llama-server -hf unsloth/gemma-3-12b-it-qat-GGUF -hff gemma-3-12b-it-qat-UD-Q4_K_XL.gguf --port ${PORT} --no-mmap --flash-attn --cache-type-k q8_0 --cache-type-v q8_0 --temp 1.0 --top-k 64 --min-p 0.00 --top-p 0.95 --repeat-penalty 1.0 --gpu-layers 49 # --no-kv-offload # --ctx-size 16384