platform/aiml/mlops/docker/serve/entrypoint.sh

#!/usr/bin/env bash
set -euo pipefail

# Config
: "${MODEL_PATH:=/data/adaptai/platform/aiml/checkpoints/qwen3-8b-elizabeth-sft}"
: "${HF_ORG:=LevelUp2x}"
: "${MODEL_NAME:=qwen3-8b-elizabeth-checkpoints}"
: "${PREFER_HF:=0}"

echo "[entrypoint] Starting Elizabeth vLLM container"
echo "[entrypoint] MODEL_PATH=${MODEL_PATH} PREFER_HF=${PREFER_HF} HF_ORG=${HF_ORG} MODEL_NAME=${MODEL_NAME}"

# Ensure caches
mkdir -p "$HF_HOME" "$HF_MODULES_CACHE" || true

# If model not present locally, try to seed
if [ ! -f "$MODEL_PATH/model.safetensors.index.json" ] && [ ! -f "$MODEL_PATH/tokenizer.json" ]; then
  echo "[entrypoint] MODEL_PATH missing expected files; attempting to seed"
  if [ -n "${SEED_HOST:-}" ]; then
    echo "[entrypoint] Seeding from SEED_HOST=$SEED_HOST"
    rsync -aH --partial --progress "${SEED_HOST}:${MODEL_PATH%/*}/" "${MODEL_PATH%/*}/" || true
  fi
fi

# If still missing and we prefer HF, download
if [ ! -f "$MODEL_PATH/model.safetensors.index.json" ] && [ "$PREFER_HF" = "1" ]; then
  if [ -z "${HF_TOKEN:-${HUGGING_FACE_API_KEY:-}}" ]; then
    echo "[entrypoint] ERROR: HF_TOKEN or HUGGING_FACE_API_KEY must be set to download from private HF repos" >&2
    exit 1
  fi
  echo "[entrypoint] Downloading from HF: ${HF_ORG}/${MODEL_NAME}"
  mkdir -p "$MODEL_PATH"
  hf download "${HF_ORG}/${MODEL_NAME}" --repo-type model --include '**' --local-dir "$MODEL_PATH" --revision main --token "${HF_TOKEN:-${HUGGING_FACE_API_KEY}}"
fi

echo "[entrypoint] Launching vLLM"
exec bash ./serve_elizabeth_vllm.sh