import numpy as np
import tensorflow as tf
from fennlp.datas.checkpoint import LoadCheckpoint
from fennlp.datas.dataloader import TFWriter, TFLoader
from fennlp.metrics import Metric
from fennlp.models import bert
# 载入参数
load_check = LoadCheckpoint(language='en')
param, vocab_file, model_path = load_check.load_bert_param()
# 定制参数
param.batch_size = 8
param.maxlen = 100
param.label_size = 9
class BERT_NER(tf.keras.Model):
def __init__(self, param, **kwargs):
super(BERT_NER, self).__init__(**kwargs)
self.batch_size = param.batch_size
self.maxlen = param.maxlen
self.label_size = param.label_size
self.bert = bert.BERT(param)
self.dense = tf.keras.layers.Dense(self.label_size, activation="relu")
def call(self, inputs, is_training=True):
bert = self.bert(inputs, is_training)
sequence_output = bert.get_sequence_output() # batch,sequence,768
from fennlp.datas.checkpoint import LoadCheckpoint
from fennlp.datas.dataloader import TFWriter, TFLoader
from fennlp.metrics import Metric
from fennlp.models import bert
from fennlp.optimizers import optim
from fennlp.tools import bert_init_weights_from_checkpoint
# 载入参数
# LoadCheckpoint(language='zh', model="bert", parameters="base", cased=True, url=None)
# language: the language you used in your input data
# model: the model you choose,could be bert albert and gpt2
# parameters: can be base large xlarge xxlarge for albert, base medium large for gpt2, base large for BERT.
# cased: True or false, only for bert model.
# url: you can give a link of other checkpoint.
load_check = LoadCheckpoint()
param, vocab_file, model_path = load_check.load_bert_param()
# 定制参数
param.batch_size = 2
param.maxlen = 4
param.label_size = 15
# 构建模型
class BERT_NER(tf.keras.Model):
def __init__(self, param, **kwargs):
super(BERT_NER, self).__init__(**kwargs)
self.batch_size = param.batch_size
self.maxlen = param.maxlen
self.label_size = param.label_size
import numpy as np
import tensorflow as tf
from fennlp.datas.checkpoint import LoadCheckpoint
from fennlp.datas.dataloader import TFWriter, TFLoader
from fennlp.metrics import Metric
from fennlp.models import albert
# 载入参数
load_check = LoadCheckpoint(language='en', model="albert", paramaters="base")
param, vocab_file, model_path, spm_model_file = load_check.load_albert_param()
# 定制参数
param.batch_size = 8
param.maxlen = 100
param.label_size = 2
# 构建模型
class ALBERT_NER(tf.keras.Model):
def __init__(self, param, **kwargs):
super(ALBERT_NER, self).__init__(**kwargs)
self.batch_size = param.batch_size
self.maxlen = param.maxlen
self.label_size = param.label_size
self.albert = albert.ALBERT(param)
self.dense = tf.keras.layers.Dense(self.label_size, activation="relu")
def call(self, inputs, is_training=True):
albert = self.albert(inputs, is_training)
sequence_output = albert.get_pooled_output() # batch,768
pre = self.dense(sequence_output)
import tensorflow as tf
from fennlp.datas.checkpoint import LoadCheckpoint
from fennlp.models import gpt2
from fennlp.sample import samples
from fennlp.tokenizers import gpt2_tokenization
from fennlp.tools import gpt2_init_weights_from_checkpoint
# 载入参数
# LoadCheckpoint(language='zh', model="bert", parameters="base", cased=True, url=None)
# language: the language you used in your input data
# model: the model you choose,could be bert albert and gpt2
# parameters: can be base large xlarge xxlarge for albert, base medium large for gpt2, base large for BERT.
# cased: True or false, only for bert model.
# url: you can give a link of other checkpoint.
load_check = LoadCheckpoint(language='en', model="gpt2", parameters="base")
param, vocab_file, model_path, encoder_file = load_check.load_gpt2_param()
print(param)
param["batch_size"] = 10
tokenizer = gpt2_tokenization.FullTokenizer(encoder_file, vocab_file)
start_token = tokenizer.convert_token_to_id("<|endoftext|>")
# 构建模型
class GenGPT2(tf.keras.Model):
def __init__(self, param, **kwargs):
super(GenGPT2, self).__init__(**kwargs)
self.model = gpt2.GPT2(param)