start_server.sh

#!/bin/bash
set -ex
echo "--- Starting setup and server ---"

python3 - <<'PY'
import sys, os
print("PYTHONPATH:", sys.path)
import modeling_dots_ocr_vllm
print("DotsOCR import OK")
PY

echo "Starting server..."
exec vllm serve ${HF_MODEL_PATH} \
    --host 0.0.0.0 \
    --port ${PORT} \
    --tensor-parallel-size 1 \
    --gpu-memory-utilization 0.8 \
    --chat-template-content-format string \
    --served-model-name dotsocr-model \
    --trust-remote-code