将自定义 BERT 问答模型转换为 Tflite 模型

Question

我有这个 BERT 模型，我对它进行了微调以回答问题，使用 SQuAD 数据集进行训练。但是，当我尝试按照指示 here 将其转换为 tflite 模型时，出现以下错误：

 'tf.Reshape' op requires 'shape' to have at most one dynamic dimension, but got multiple dynamic dimensions at indices 0 and 1
<unknown>:0: note: loc("StatefulPartitionedCall"): 

以下是转换代码

import tensorflow as tf

saved_model_dir = '/home/vineetkalghatgi/Downloads/mymodel-20210413T165017Z-001/mymodel'

# Convert the model
converter = tf.lite.TFLiteConverter.from_saved_model(saved_model_dir) # path to the SavedModel directory
tflite_model = None
try:
  tflite_model = converter.convert()
except Exception as e:
  print(e)
  exit()

# Save the model.
with open('model.tflite', 'wb') as f:
  f.write(tflite_model)