Add Qwen2.5 tokenizer compatibility engine and runner workflow

vllm_wrapper.sh

@@ -0,0 +1,25 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+vllm_real="$(dirname "$(readlink -f "$0")")/vllm_real"
+
+if [[ "${1:-}" != "serve" || -z "${2:-}" ]]; then
+    exec "$vllm_real" "$@"
+fi
+
+model_dir="$2"
+shift 2
+prepared="$(python3 /opt/modelhub/engine_compat.py "$model_dir")"
+if [[ -z "$prepared" ]]; then
+    exec "$vllm_real" serve "$model_dir" "$@"
+fi
+
+args=("$@")
+for ((i=0; i<${#args[@]}; i++)); do
+    if [[ "${args[i]}" == "--tokenizer" && $((i+1)) -lt ${#args[@]} && "${args[i+1]}" == "$model_dir" ]]; then
+        args[i+1]="$prepared"
+    elif [[ "${args[i]}" == "--tokenizer=$model_dir" ]]; then
+        args[i]="--tokenizer=$prepared"
+    fi
+done
+exec "$vllm_real" serve "$prepared" "${args[@]}"