start.sh

#!/bin/bash

# Start llama.cpp server in background
/app/llama.cpp/build/bin/llama-server \
    -m /app/models/${ORPHEUS_MODEL_NAME} \
    --host 0.0.0.0 \
    --port 5006 \
    --ctx-size ${ORPHEUS_MAX_TOKENS:-512} \
    --n-predict ${ORPHEUS_MAX_TOKENS:-512} \
    --threads ${LLAMA_CPU_THREADS:-6} \
    --threads-batch ${LLAMA_CPU_THREADS:-6} \
    --rope-scaling linear \
    --no-mmap \
    --no-slots &

# Wait briefly to allow server to start
sleep 5

# Start FastAPI server with logging enabled
uvicorn orpheus.app:app \
    --host 0.0.0.0 \
    --port 5005 \
    --log-level info