导入ktrain时遇tensorflow.keras无swish属性报错的解决方法

问题描述 #

导入ktrain库时触发报错：AttributeError: module 'tensorflow_core.keras.activations' has no attribute 'swish'，相同代码在Google Colab环境中可以正常运行，完整报错堆栈如下：

AttributeError                            Traceback (most recent call last)
<ipython-input-3-d851226d5f91> in <module>
----> 1 import ktrain
      2 from ktrain import text

~/opt/anaconda3/envs/Anmol/lib/python3.7/site-packages/ktrain/__init__.py in <module>
      1 from .version import __version__
----> 2 from . import imports as I
      3 from .core import ArrayLearner, GenLearner, get_predictor, load_predictor, release_gpu_memory
      4 from .vision.learner import ImageClassLearner
      5 from .text.learner import BERTTextClassLearner, TransformerTextClassLearner

~/opt/anaconda3/envs/Anmol/lib/python3.7/site-packages/ktrain/imports.py in <module>
    227 logging.getLogger("transformers").setLevel(logging.ERROR)
    228 try:
--> 229     import transformers
    230 except ImportError:
    231     warnings.warn("transformers not installed - needed by various models in 'text' module")

~/opt/anaconda3/envs/Anmol/lib/python3.7/site-packages/transformers/__init__.py in <module>
    133 
    134 # Pipelines
--> 135 from .pipelines import (
    136     Conversation,
    137     ConversationalPipeline,

~/opt/anaconda3/envs/Anmol/lib/python3.7/site-packages/transformers/pipelines.py in <module>
     46     import tensorflow as tf
     47 
---> 48     from .modeling_tf_auto import (
     49         TF_MODEL_FOR_QUESTION_ANSWERING_MAPPING,
     50         TF_MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING,

~/opt/anaconda3/envs/Anmol/lib/python3.7/site-packages/transformers/modeling_tf_auto.py in <module>
     49 from .configuration_utils import PretrainedConfig
     50 from .file_utils import add_start_docstrings
---> 51 from .modeling_tf_albert import (
     52     TFAlbertForMaskedLM,
     53     TFAlbertForMultipleChoice,

~/opt/anaconda3/envs/Anmol/lib/python3.7/site-packages/transformers/modeling_tf_albert.py in <module>
     22 import tensorflow as tf
     23 
---> 24 from .activations_tf import get_tf_activation
     25 from .configuration_albert import AlbertConfig
     26 from .file_utils import (

~/opt/anaconda3/envs/Anmol/lib/python3.7/site-packages/transformers/activations_tf.py in <module>
     52     "gelu": tf.keras.layers.Activation(gelu),
     53     "relu": tf.keras.activations.relu,
---> 54     "swish": tf.keras.activations.swish,
     55     "silu": tf.keras.activations.swish,
     56     "gelu_new": tf.keras.layers.Activation(gelu_new),

AttributeError: module 'tensorflow_core.keras.activations' has no attribute 'swish'

报错根因 #

错误并非ktrain本身抛出，触发点在transformers库的加载流程：swish激活函数是TensorFlow 2.2版本才正式集成到keras.activations模块的，你本地conda环境的TensorFlow版本低于2.2，本身不存在这个属性，而当前安装的高版本transformers直接调用了这个不存在的属性，就会触发报错。Google Colab的依赖版本经过兼容性匹配，所以不会出现同类问题。

解决步骤 #

• 先核查本地TensorFlow版本，执行命令：pip show tensorflow，确认当前版本号
• 不改动现有TF版本的前提下，直接把transformers降级到适配低版本TF的兼容版本，执行命令：pip install transformers==4.10.0，这个版本不会强制依赖仅高版本TF才有的swish属性
• 如果可以升级TF版本，直接安装2.6及以上的稳定版TF（适配Python3.7环境），执行命令：pip install --upgrade tensorflow==2.6.5，安装完成后运行验证代码import tensorflow as tf; print(tf.keras.activations.swish)，无报错即说明属性可正常调用
• 补充排查：如果是M系列芯片的Mac搭建的conda环境，检查是否存在旧版tensorflow-macos残留包，残留包会导致keras模块路径错乱，需将所有TF相关旧包卸载干净后再重装适配版本

内容的提问来源于stack exchange，提问作者Anmol Gupta