local-llm-server/llm_server/routes/v1/proxy.py

from flask import jsonify

from llm_server.custom_redis import flask_cache
from . import bp
from .generate_stats import generate_stats
from ..auth import requires_auth
from ...cluster.backend import get_backends
from ...cluster.cluster_config import cluster_config
from ...helpers import jsonify_pretty


@bp.route('/stats', methods=['GET'])
@flask_cache.cached(timeout=5, query_string=True)
def get_stats():
    return jsonify_pretty(generate_stats())


@bp.route('/backends', methods=['GET'])
@requires_auth
def get_backend():
    online, offline = get_backends()
    result = {}
    for i in online + offline:
        info = cluster_config.get_backend(i)
        result[info['hash']] = info
    return jsonify(result), 200