mAP drop rapidly when training

When I train yolox-s model on my own dataset, the mAP drop rapidly on the 45 epoch.
like this:

Is this a kind of normal situation?

My exp file:

#!/usr/bin/env python3
# -*- coding:utf-8 -*-
# Copyright (c) Megvii, Inc. and its affiliates.
import os

from yolox.exp import Exp as MyExp


class Exp(MyExp):
    def __init__(self):
        super(Exp, self).__init__()
        self.depth = 0.33
        self.width = 0.50
        self.exp_name = os.path.split(os.path.realpath(__file__))[1].split(".")[0]

        # Define yourself dataset path
        self.data_dir = "datasets/my_coco_dataset"
        self.train_ann = "instances_train2017.json"
        self.val_ann = "instances_val2017.json"

        self.num_classes = 9

        self.max_epoch = 200
        self.data_num_workers = 4
        self.eval_interval = 1

        # self.basic_lr_per_img = 0.01 / 128.0  # 0.01 / 64.0
        # self.min_lr_ratio = 0.025  # 0.05

My training script:

python tools/train.py -f exps/example/custom/yolox_s.py -expn my_yoloxs_8device_200epoch_64batch_640 -d 8 -b 64 -o

Start with tools/train.py and the supplied exps/example/custom/yolox_s.py configuration, then review how evaluation is run each epoch and how the mAP result is recorded around epoch 45. Done means determining whether the reported drop is expected for this configuration or producing a reproducible diagnosis from the provided dataset and training settings.