diff --git a/douzero/server/battle.py b/douzero/server/battle.py
index c34340c..70c720d 100644
--- a/douzero/server/battle.py
+++ b/douzero/server/battle.py
@@ -2,10 +2,25 @@ import os
 import traceback
 
 from .orm import Baseline, Battle, Model
-from evaluate import evaluate
+from ..evaluation.simulation import evaluate
+from ..evaluation.deep_agent import DeepAgent
 from datetime import datetime
 
 positions = ['landlord', 'landlord_up', 'landlord_front', 'landlord_down']
+idx_position = {0: 'landlord', 1: 'landlord_down', 2: 'landlord_front', 3:'landlord_up'}
+position_idx = {'landlord': 0, 'landlord_down': 1, 'landlord_front': 2, 'landlord_up': 3}
+
+baselines = Baseline.select().order_by(Baseline.rank.desc()).limit(1)
+baseline_players = [None, None, None, None]
+if len(baselines) >= 1:
+    baseline = baselines[0]
+    try:
+        baseline_players[0] = DeepAgent('landlord', str(baseline.landlord_path), use_onnx=True)
+        baseline_players[1] = DeepAgent('landlord_down', str(baseline.landlord_down_path), use_onnx=True)
+        baseline_players[2] = DeepAgent('landlord_front', str(baseline.landlord_front_path), use_onnx=True)
+        baseline_players[3] = DeepAgent('landlord_up', str(baseline.landlord_up_path), use_onnx=True)
+    except:
+        pass
 
 def battle_logic(baseline : Baseline, battle : Battle):
     eval_data_first = 'eval_data_200.pkl'
@@ -77,6 +92,8 @@ def battle_logic(baseline : Baseline, battle : Battle):
         if os.path.exists(onnx_path):
             os.remove(onnx_path)
         os.remove(str(battle.challenger_path))
+    else:
+        baseline_players[position_idx[battle.challenger_position]] = DeepAgent(battle.challenger_position, str(battle.challenger_path), use_onnx=True)
 
 def tick():
     try:
@@ -91,6 +108,7 @@ def tick():
                     models = Model.select().where(Model.position == position).order_by(Model.create_time.desc()).limit(1)
                     if(len(models) > 0):
                         baseline['%s_path' % position] = models[0].path
+                        baseline_players[position_idx[position]] = DeepAgent(position, str(models[0].path), use_onnx=True)
                 if len(baseline.keys()) == 4:
                     baseline['rank'] = 0
                     baseline['landlord_wp'] = 0
diff --git a/evaluate_server.py b/evaluate_server.py
index 4460880..68f3159 100644
--- a/evaluate_server.py
+++ b/evaluate_server.py
@@ -4,14 +4,13 @@ import os
 from peewee import JOIN
 
 from douzero.server.orm import Model, Battle, Baseline
-from douzero.server.battle import tick
+from douzero.server.battle import tick, baseline_players, positions, idx_position
 from flask import Flask, jsonify, request
 from flask_cors import CORS
 from datetime import datetime
 from concurrent.futures import ThreadPoolExecutor
 
 from douzero.env.game import get_legal_card_play_actions
-from douzero.evaluation.deep_agent import DeepAgent
 
 app = Flask(__name__)
 CORS(app)
@@ -19,8 +18,6 @@ CORS(app)
 Model.create_table()
 Battle.create_table()
 Baseline.create_table()
-positions = ['landlord', 'landlord_up', 'landlord_front', 'landlord_down']
-idx_position = {0: 'landlord', 1: 'landlord_down', 2: 'landlord_front', 3:'landlord_up'}
 
 threadpool = ThreadPoolExecutor(1)
 
@@ -33,18 +30,6 @@ RealCard2EnvCard = {'3': 3, '4': 4, '5': 5, '6': 6, '7': 7,
                     '8': 8, '9': 9, 'T': 10, 'J': 11, 'Q': 12,
                     'K': 13, 'A': 14, '2': 17, 'X': 20, 'D': 30}
 
-baselines = Baseline.select().order_by(Baseline.rank.desc()).limit(1)
-if len(baselines) >= 1:
-    baseline = baselines[0]
-    try:
-        players = [
-            DeepAgent('landlord', str(baseline.landlord_path), use_onnx=True),
-            DeepAgent('landlord_down', str(baseline.landlord_down_path), use_onnx=True),
-            DeepAgent('landlord_front', str(baseline.landlord_front_path), use_onnx=True),
-            DeepAgent('landlord_up', str(baseline.landlord_up_path), use_onnx=True)
-        ]
-    except:
-        pass
 
 @app.route('/upload', methods=['POST'])
 def upload():
@@ -211,7 +196,7 @@ def predict():
             info_set.legal_actions = _get_legal_card_play_actions(player_hand_cards, rival_move)
 
             # Prediction
-            actions, actions_confidence = players[player_position].act(info_set, True)
+            actions, actions_confidence = baseline_players[player_position].act(info_set, True)
             actions = [''.join([EnvCard2RealCard[a] for a in action]) for action in actions]
             result = {}
             win_rates = {}