开发笔记:elmo模型

import tensorflow_hub as hub
# 加载模型
elmo = hub.Module("https://tfhub.dev/google/elmo/2", trainable=True)
# 输入的数据集
texts = ["the cat is on the mat", "dogs are in the fog"]
embeddings = elmo(
texts,
signature="default",
as_dict=True)["default"]
elmo = hub.Module("https://tfhub.dev/google/elmo/2", trainable=True)
# 另一种方式输入数据
tokens_input = [["the", "cat", "is", "on", "the", "mat"],
["dogs", "are", "in", "the", "fog", ""]]
# 长度，表示tokens_input第一行6一个有效，第二行5个有效
tokens_length = [6, 5]
# 生成elmo embedding
embeddings = elmo(
inputs={
"tokens": tokens_input,
"sequence_len": tokens_length
},
signature="tokens",
as_dict=True)["default"]
from tensorflow.python.keras import backend as K
sess = K.get_session()
array = sess.run(embeddings)