kashgar_chinese_ner.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 357
    },
    "colab_type": "code",
    "id": "MHXIom1tFv3E",
    "outputId": "b551c791-7db9-49f8-c317-48ef505049eb"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "--2019-05-30 06:34:33--  https://storage.googleapis.com/bert_models/2018_11_03/chinese_L-12_H-768_A-12.zip\n",
      "Resolving storage.googleapis.com (storage.googleapis.com)... 172.217.212.128, 2607:f8b0:4001:c05::80\n",
      "Connecting to storage.googleapis.com (storage.googleapis.com)|172.217.212.128|:443... connected.\n",
      "HTTP request sent, awaiting response... 200 OK\n",
      "Length: 381892918 (364M) [application/zip]\n",
      "Saving to: ‘chinese_L-12_H-768_A-12.zip’\n",
      "\n",
      "chinese_L-12_H-768_ 100%[===================>] 364.20M   105MB/s    in 3.5s    \n",
      "\n",
      "2019-05-30 06:34:37 (105 MB/s) - ‘chinese_L-12_H-768_A-12.zip’ saved [381892918/381892918]\n",
      "\n",
      "Archive:  chinese_L-12_H-768_A-12.zip\n",
      "   creating: chinese_L-12_H-768_A-12/\n",
      "  inflating: chinese_L-12_H-768_A-12/bert_model.ckpt.meta  \n",
      "  inflating: chinese_L-12_H-768_A-12/bert_model.ckpt.data-00000-of-00001  \n",
      "  inflating: chinese_L-12_H-768_A-12/vocab.txt  \n",
      "  inflating: chinese_L-12_H-768_A-12/bert_model.ckpt.index  \n",
      "  inflating: chinese_L-12_H-768_A-12/bert_config.json  \n",
      "bert_config.json\t\t     bert_model.ckpt.index  vocab.txt\n",
      "bert_model.ckpt.data-00000-of-00001  bert_model.ckpt.meta\n"
     ]
    }
   ],
   "source": [
    "# Download BERT\n",
    "!wget https://storage.googleapis.com/bert_models/2018_11_03/chinese_L-12_H-768_A-12.zip\n",
    "!unzip chinese_L-12_H-768_A-12.zip\n",
    "!ls chinese_L-12_H-768_A-12"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "mKVBZdCrDacZ"
   },
   "outputs": [],
   "source": [
    "!pip install kashgari\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "r3mGKW2Q7mli",
    "outputId": "d963ba56-49af-4454-e16c-58af6152b56b"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "chinese_L-12_H-768_A-12\n"
     ]
    }
   ],
   "source": [
    "# BERT_PATH = '/input0/BERT/chinese_L-12_H-768_A-12/'\n",
    "BERT_PATH = 'chinese_L-12_H-768_A-12'\n",
    "print(BERT_PATH)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 85
    },
    "colab_type": "code",
    "id": "bkrSeQSI7pTT",
    "outputId": "c4d3052e-9197-46d0-89c8-e82fc33f4b72"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Creating data folder...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "file_sizes:   0%|                                   | 0.00/2.44M [00:00<?, ?B/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading data from http://storage.eliyar.biz/corpus/china-people-daily-ner-corpus.tar.gz (2.3 MB)\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "file_sizes: 100%|██████████████████████████| 2.44M/2.44M [00:02<00:00, 1.10MB/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting tar.gz file...\n",
      "Successfully downloaded / unzipped to /root/.kashgari/corpus\n",
      "train data count: 20864\n",
      "validate data count: 2318\n",
      "test data count: 4636\n"
     ]
    }
   ],
   "source": [
    "from kashgari.corpus import ChinaPeoplesDailyNerCorpus\n",
    "\n",
    "train_x, train_y = ChinaPeoplesDailyNerCorpus.get_sequence_tagging_data('train')\n",
    "validate_x, validate_y = ChinaPeoplesDailyNerCorpus.get_sequence_tagging_data('validate')\n",
    "test_x, test_y  = ChinaPeoplesDailyNerCorpus.get_sequence_tagging_data('test')\n",
    "\n",
    "print(f\"train data count: {len(train_x)}\")\n",
    "print(f\"validate data count: {len(validate_x)}\")\n",
    "print(f\"test data count: {len(test_x)}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 221
    },
    "colab_type": "code",
    "id": "xplap8lL7t2b",
    "outputId": "620b940a-6747-47c4-f3f5-23b99cdf863a"
   },
   "outputs": [],
   "source": [
    "from kashgari.embeddings import BERTEmbedding\n",
    "embedding = BERTEmbedding(BERT_PATH, 100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "HmeueeKY7vv5"
   },
   "outputs": [],
   "source": [
    "from kashgari.tasks.seq_labeling import BLSTMCRFModel"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = BLSTMCRFModel(embedding)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 4267
    },
    "colab_type": "code",
    "id": "_mzlD5GeCpUs",
    "outputId": "031241ae-b37d-4138-f042-55575b23d898"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/50\n",
      "20/20 [==============================] - 66s 3s/step - loss: -0.0018 - crf_accuracy: 0.9991 - val_loss: 0.0058 - val_crf_accuracy: 0.9965\n",
      "Epoch 2/50\n",
      "20/20 [==============================] - 62s 3s/step - loss: -0.0023 - crf_accuracy: 0.9993 - val_loss: 8.0986e-04 - val_crf_accuracy: 0.9977\n",
      "Epoch 3/50\n",
      "20/20 [==============================] - 66s 3s/step - loss: -0.0025 - crf_accuracy: 0.9993 - val_loss: 0.0055 - val_crf_accuracy: 0.9966\n",
      "Epoch 4/50\n",
      "20/20 [==============================] - 63s 3s/step - loss: -0.0028 - crf_accuracy: 0.9994 - val_loss: 0.0047 - val_crf_accuracy: 0.9968\n",
      "Epoch 5/50\n",
      "20/20 [==============================] - 66s 3s/step - loss: -0.0029 - crf_accuracy: 0.9994 - val_loss: 0.0057 - val_crf_accuracy: 0.9965\n",
      "Epoch 6/50\n",
      "20/20 [==============================] - 65s 3s/step - loss: -0.0030 - crf_accuracy: 0.9994 - val_loss: 0.0049 - val_crf_accuracy: 0.9967\n",
      "Epoch 7/50\n",
      "20/20 [==============================] - 66s 3s/step - loss: -0.0033 - crf_accuracy: 0.9995 - val_loss: 0.0054 - val_crf_accuracy: 0.9966\n",
      "Epoch 8/50\n",
      "20/20 [==============================] - 63s 3s/step - loss: -0.0037 - crf_accuracy: 0.9996 - val_loss: 0.0038 - val_crf_accuracy: 0.9970\n",
      "Epoch 9/50\n",
      "20/20 [==============================] - 63s 3s/step - loss: -0.0038 - crf_accuracy: 0.9996 - val_loss: 0.0049 - val_crf_accuracy: 0.9967\n",
      "Epoch 10/50\n",
      "20/20 [==============================] - 65s 3s/step - loss: -0.0039 - crf_accuracy: 0.9995 - val_loss: 0.0042 - val_crf_accuracy: 0.9967\n",
      "Epoch 11/50\n",
      "20/20 [==============================] - 62s 3s/step - loss: -0.0042 - crf_accuracy: 0.9996 - val_loss: 0.0033 - val_crf_accuracy: 0.9970\n",
      "Epoch 12/50\n",
      "20/20 [==============================] - 66s 3s/step - loss: -0.0044 - crf_accuracy: 0.9996 - val_loss: 0.0049 - val_crf_accuracy: 0.9967\n",
      "Epoch 13/50\n",
      "20/20 [==============================] - 66s 3s/step - loss: -0.0045 - crf_accuracy: 0.9996 - val_loss: 0.0033 - val_crf_accuracy: 0.9970\n",
      "Epoch 14/50\n",
      "20/20 [==============================] - 64s 3s/step - loss: -0.0046 - crf_accuracy: 0.9996 - val_loss: 0.0043 - val_crf_accuracy: 0.9968\n",
      "Epoch 15/50\n",
      "20/20 [==============================] - 66s 3s/step - loss: -0.0048 - crf_accuracy: 0.9997 - val_loss: 0.0052 - val_crf_accuracy: 0.9967\n",
      "Epoch 16/50\n",
      "20/20 [==============================] - 64s 3s/step - loss: -0.0050 - crf_accuracy: 0.9997 - val_loss: 0.0041 - val_crf_accuracy: 0.9968\n",
      "Epoch 17/50\n",
      "20/20 [==============================] - 62s 3s/step - loss: -0.0052 - crf_accuracy: 0.9997 - val_loss: 0.0031 - val_crf_accuracy: 0.9970\n",
      "Epoch 18/50\n",
      "20/20 [==============================] - 67s 3s/step - loss: -0.0053 - crf_accuracy: 0.9997 - val_loss: 0.0045 - val_crf_accuracy: 0.9967\n",
      "Epoch 19/50\n",
      "20/20 [==============================] - 66s 3s/step - loss: -0.0054 - crf_accuracy: 0.9997 - val_loss: 0.0050 - val_crf_accuracy: 0.9966\n",
      "Epoch 20/50\n",
      "20/20 [==============================] - 64s 3s/step - loss: -0.0056 - crf_accuracy: 0.9997 - val_loss: 0.0029 - val_crf_accuracy: 0.9970\n",
      "Epoch 21/50\n",
      "20/20 [==============================] - 64s 3s/step - loss: -0.0058 - crf_accuracy: 0.9998 - val_loss: 0.0039 - val_crf_accuracy: 0.9970\n",
      "Epoch 22/50\n",
      "20/20 [==============================] - 64s 3s/step - loss: -0.0059 - crf_accuracy: 0.9998 - val_loss: 0.0039 - val_crf_accuracy: 0.9969\n",
      "Epoch 23/50\n",
      "20/20 [==============================] - 64s 3s/step - loss: -0.0061 - crf_accuracy: 0.9998 - val_loss: 0.0026 - val_crf_accuracy: 0.9971\n",
      "Epoch 24/50\n",
      "20/20 [==============================] - 65s 3s/step - loss: -0.0062 - crf_accuracy: 0.9998 - val_loss: 0.0041 - val_crf_accuracy: 0.9968\n",
      "Epoch 25/50\n",
      "20/20 [==============================] - 66s 3s/step - loss: -0.0064 - crf_accuracy: 0.9998 - val_loss: 0.0043 - val_crf_accuracy: 0.9967\n",
      "Epoch 26/50\n",
      "20/20 [==============================] - 63s 3s/step - loss: -0.0064 - crf_accuracy: 0.9998 - val_loss: -0.0016 - val_crf_accuracy: 0.9976\n",
      "Epoch 27/50\n",
      "20/20 [==============================] - 68s 3s/step - loss: -0.0065 - crf_accuracy: 0.9997 - val_loss: 0.0040 - val_crf_accuracy: 0.9967\n",
      "Epoch 28/50\n",
      "20/20 [==============================] - 63s 3s/step - loss: -0.0068 - crf_accuracy: 0.9998 - val_loss: 0.0036 - val_crf_accuracy: 0.9967\n",
      "Epoch 29/50\n",
      "20/20 [==============================] - 67s 3s/step - loss: -0.0069 - crf_accuracy: 0.9998 - val_loss: 0.0041 - val_crf_accuracy: 0.9966\n",
      "Epoch 30/50\n",
      "20/20 [==============================] - 62s 3s/step - loss: -0.0072 - crf_accuracy: 0.9998 - val_loss: 0.0031 - val_crf_accuracy: 0.9968\n",
      "Epoch 31/50\n",
      "20/20 [==============================] - 66s 3s/step - loss: -0.0073 - crf_accuracy: 0.9998 - val_loss: 0.0039 - val_crf_accuracy: 0.9967\n",
      "Epoch 32/50\n",
      "20/20 [==============================] - 66s 3s/step - loss: -0.0074 - crf_accuracy: 0.9998 - val_loss: 0.0014 - val_crf_accuracy: 0.9972\n",
      "Epoch 33/50\n",
      "20/20 [==============================] - 65s 3s/step - loss: -0.0076 - crf_accuracy: 0.9998 - val_loss: 0.0026 - val_crf_accuracy: 0.9970\n",
      "Epoch 34/50\n",
      "20/20 [==============================] - 64s 3s/step - loss: -0.0077 - crf_accuracy: 0.9998 - val_loss: 0.0031 - val_crf_accuracy: 0.9969\n",
      "Epoch 35/50\n",
      "20/20 [==============================] - 63s 3s/step - loss: -0.0078 - crf_accuracy: 0.9998 - val_loss: -0.0026 - val_crf_accuracy: 0.9980\n",
      "Epoch 36/50\n",
      "20/20 [==============================] - 66s 3s/step - loss: -0.0080 - crf_accuracy: 0.9998 - val_loss: 0.0037 - val_crf_accuracy: 0.9966\n",
      "Epoch 37/50\n",
      "20/20 [==============================] - 63s 3s/step - loss: -0.0082 - crf_accuracy: 0.9998 - val_loss: 0.0020 - val_crf_accuracy: 0.9971\n",
      "Epoch 38/50\n",
      "20/20 [==============================] - 66s 3s/step - loss: -0.0083 - crf_accuracy: 0.9998 - val_loss: 0.0042 - val_crf_accuracy: 0.9967\n",
      "Epoch 39/50\n",
      "20/20 [==============================] - 64s 3s/step - loss: -0.0085 - crf_accuracy: 0.9999 - val_loss: 0.0019 - val_crf_accuracy: 0.9971\n",
      "Epoch 40/50\n",
      "20/20 [==============================] - 64s 3s/step - loss: -0.0086 - crf_accuracy: 0.9998 - val_loss: 0.0012 - val_crf_accuracy: 0.9971\n",
      "Epoch 41/50\n",
      "20/20 [==============================] - 65s 3s/step - loss: -0.0087 - crf_accuracy: 0.9998 - val_loss: 0.0051 - val_crf_accuracy: 0.9964\n",
      "Epoch 42/50\n",
      "20/20 [==============================] - 65s 3s/step - loss: -0.0089 - crf_accuracy: 0.9999 - val_loss: 0.0015 - val_crf_accuracy: 0.9971\n",
      "Epoch 43/50\n",
      "20/20 [==============================] - 64s 3s/step - loss: -0.0090 - crf_accuracy: 0.9998 - val_loss: 0.0025 - val_crf_accuracy: 0.9969\n",
      "Epoch 44/50\n",
      "20/20 [==============================] - 64s 3s/step - loss: -0.0092 - crf_accuracy: 0.9999 - val_loss: 3.8593e-04 - val_crf_accuracy: 0.9971\n",
      "Epoch 45/50\n",
      "20/20 [==============================] - 65s 3s/step - loss: -0.0094 - crf_accuracy: 0.9999 - val_loss: 0.0027 - val_crf_accuracy: 0.9968\n",
      "Epoch 46/50\n",
      "20/20 [==============================] - 64s 3s/step - loss: -0.0095 - crf_accuracy: 0.9999 - val_loss: -1.1398e-04 - val_crf_accuracy: 0.9971\n",
      "Epoch 47/50\n",
      "20/20 [==============================] - 66s 3s/step - loss: -0.0097 - crf_accuracy: 0.9999 - val_loss: 0.0024 - val_crf_accuracy: 0.9967\n",
      "Epoch 48/50\n",
      "20/20 [==============================] - 64s 3s/step - loss: -0.0099 - crf_accuracy: 0.9999 - val_loss: 0.0012 - val_crf_accuracy: 0.9970\n",
      "Epoch 49/50\n",
      "20/20 [==============================] - 64s 3s/step - loss: -0.0100 - crf_accuracy: 0.9999 - val_loss: 0.0011 - val_crf_accuracy: 0.9971\n",
      "Epoch 50/50\n",
      "20/20 [==============================] - 65s 3s/step - loss: -0.0101 - crf_accuracy: 0.9999 - val_loss: 0.0021 - val_crf_accuracy: 0.9968\n"
     ]
    }
   ],
   "source": [
    "model.fit(train_x,\n",
    "          train_y,\n",
    "          x_validate=validate_x,\n",
    "          y_validate=validate_y,\n",
    "          epochs=100,\n",
    "          batch_size=1024)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 207
    },
    "colab_type": "code",
    "id": "hb_6hCMcn7UK",
    "outputId": "9071ea6e-532f-42df-a12e-d9f1a0dda460"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "           precision    recall  f1-score   support\n",
      "\n",
      "      LOC     0.9296    0.9359    0.9328      3431\n",
      "      PER     0.9629    0.9666    0.9647      1797\n",
      "      ORG     0.8680    0.8761    0.8720      2147\n",
      "\n",
      "micro avg     0.9197    0.9260    0.9228      7375\n",
      "macro avg     0.9198    0.9260    0.9229      7375\n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "'           precision    recall  f1-score   support\\n\\n      LOC     0.9296    0.9359    0.9328      3431\\n      PER     0.9629    0.9666    0.9647      1797\\n      ORG     0.8680    0.8761    0.8720      2147\\n\\nmicro avg     0.9197    0.9260    0.9228      7375\\nmacro avg     0.9198    0.9260    0.9229      7375\\n'"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.evaluate(test_x, test_y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "nCLC7TWTHWPG"
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "collapsed_sections": [],
   "include_colab_link": true,
   "name": "Kashgar_Chinese_NER",
   "provenance": [],
   "version": "0.3.2"
  },
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}

你好，请问模型保存和载入的接口是什么啊？

你好，请问模型保存和载入的接口是什么啊？

请看这里 https://github.com/BrikerMan/Kashgari/

请问如何指定GPU训练？

请问如何指定GPU训练？

只要安装 tensorflow-gpu 就可以，更多请看 https://github.com/BrikerMan/Kashgari/ 的文档

请问训练好模型用怎么predict呢，具体的代码是什么

可以使用 predict 或者 predict_entities API， 文档路径：https://kashgari.bmio.net/api/tasks.labeling/#predict ，建议使用新版本。

出现错误:ZeroDivisionError: Weights sum to zero, can't be normalized是什么原因呢？

这个笔记本已经过期，请按照 https://kashgari.bmio.net 重新写一个测试。

您好 请问能不能使用自己的实体标签？

您好 请问能不能使用自己的实体标签？

可以的，请看这里 https://kashgari-zh.bmio.net/tutorial/text-labeling/#_2

谢谢，已经成功了，但是这个预测是不是很慢啊？我打印了下时间2457条数据预测时间花费了511秒，请问这正常吗？

BERT 预测就是慢，可以通过 GPU + TF-Serving 部署方式稍微提高一些。

您好 请问能不能使用自己的实体标签？

可以的，请看这里 https://kashgari-zh.bmio.net/tutorial/text-labeling/#_2

我想用自己的数据集，并且标注的也不是原来的PER等，也是自己定义的实体标签，数据格式一致后，不用指定实体标签有哪些，程序会自动识别有哪些实体标签是么？

有

对的，你可以指定自己的标签，程序会自动处理。可以参考这里

https://github.com/BrikerMan/Kashgari/blob/8831993ff32efd91ab1689b864f7037310670e84/tests/corpus.py#L61

有

对的，你可以指定自己的标签，程序会自动处理。可以参考这里

https://github.com/BrikerMan/Kashgari/blob/8831993ff32efd91ab1689b864f7037310670e84/tests/corpus.py#L61

好的，谢谢您啦，我先自己研究下，后面有问题了再请教您！再次感谢

有

对的，你可以指定自己的标签，程序会自动处理。可以参考这里

https://github.com/BrikerMan/Kashgari/blob/8831993ff32efd91ab1689b864f7037310670e84/tests/corpus.py#L61

您好，请问预测出现：（1）实体没有以‘’B-xx"开头，直接就是“I-xx”；（2）实体的多个字的“I-xxx”中间出现“O”截断；（3）实体以“B-labelA”开头，后面却出现“I-labelB”
不知道一般是由什么原因导致的呢？用的是BiLSTM_CRF_Model，为什么会出现这种情况呢？

你好，有一个简单的问题。 调动模型的fit方法，训练好模型后，如何保存模型呢？
以后做预测的时候是用 predict方法直接进行预测吗？

你好，有一个简单的问题。 调动模型的fit方法，训练好模型后，如何保存模型呢？
以后做预测的时候是用 predict方法直接进行预测吗？

是的，详细使用请参考 https://kashgari-zh.bmio.net

你好，请问可以使用BMESO标注的数据吗，还是只能使用BIO标注的数据呢

目前只测试过 BIO 格式，你也可以试试别的格式。不过建议 BIO。 hziheng <notifications@github.com>于2019年9月25日 周三19:33写道：

…

你好，请问可以使用BMESO标注的数据吗，还是只能使用BIO标注的数据呢 — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <https://gist.github.com/efd8c95b0450fc73f387e0fc85ad7408?email_source=notifications&email_token=ACHPKS3D47RALMUSWGNS5YLQLNECBA5CNFSM4H5DKBF2YY3PNVWWK3TUL52HS4DFVNDWS43UINXW23LFNZ2KUY3PNVWWK3TUL5UWJTQAFZL2A#gistcomment-3037088>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACHPKS6K3WASINONUUS74JLQLNECBANCNFSM4H5DKBFQ> .

请问在In [3]:中报错 timeout 可能是啥原因？谢谢

可以选定多gpu训练吗

tf2.0的版本bert+bilstm+crf出了吗？