Python modeling.get_assignment_map_from_checkpoint方法代码示例

# 需要导入模块: from bert import modeling [as 别名]
# 或者: from bert.modeling import get_assignment_map_from_checkpoint [as 别名]
def _restore_checkpoint(init_checkpoint):
  """Restore parameters from checkpoint."""
  tvars = tf.trainable_variables()
  (assignment_map,
   _) = modeling.get_assignment_map_from_checkpoint(tvars, init_checkpoint)
  tf.train.init_from_checkpoint(init_checkpoint, assignment_map) 

# 需要导入模块: from bert import modeling [as 别名]
# 或者: from bert.modeling import get_assignment_map_from_checkpoint [as 别名]
def get_assignment_map_from_checkpoint(tvars, init_checkpoint,
                                       load_only_bert=False):
  """Compute the union of the current variables and checkpoint variables."""
  assignment_map = {}
  initialized_variable_names = {}

  name_to_variable = collections.OrderedDict()
  for var in tvars:
    name = var.name
    m = re.match("^(.*):\\d+$", name)
    if m is not None:
      name = m.group(1)
    if load_only_bert and ("bert" not in name):
      continue
    name_to_variable[name] = var

  init_vars = tf.train.list_variables(init_checkpoint)

  assignment_map = collections.OrderedDict()
  for x in init_vars:
    (name, var) = (x[0], x[1])
    if name not in name_to_variable:
      continue
    assignment_map[name] = name
    initialized_variable_names[name] = 1
    initialized_variable_names[name + ":0"] = 1

  return (assignment_map, initialized_variable_names) 

# 需要导入模块: from bert import modeling [as 别名]
# 或者: from bert.modeling import get_assignment_map_from_checkpoint [as 别名]
def __init__(self, config):
    self.config = config
    self.max_segment_len = config['max_segment_len']
    self.max_span_width = config["max_span_width"]
    self.genres = { g:i for i,g in enumerate(config["genres"]) }
    self.subtoken_maps = {}
    self.gold = {}
    self.eval_data = None # Load eval data lazily.
    self.bert_config = modeling.BertConfig.from_json_file(config["bert_config_file"])
    self.tokenizer = tokenization.FullTokenizer(
                vocab_file=config['vocab_file'], do_lower_case=False)

    input_props = []
    input_props.append((tf.int32, [None, None])) # input_ids.
    input_props.append((tf.int32, [None, None])) # input_mask
    input_props.append((tf.int32, [None])) # Text lengths.
    input_props.append((tf.int32, [None, None])) # Speaker IDs.
    input_props.append((tf.int32, [])) # Genre.
    input_props.append((tf.bool, [])) # Is training.
    input_props.append((tf.int32, [None])) # Gold starts.
    input_props.append((tf.int32, [None])) # Gold ends.
    input_props.append((tf.int32, [None])) # Cluster ids.
    input_props.append((tf.int32, [None])) # Sentence Map

    self.queue_input_tensors = [tf.placeholder(dtype, shape) for dtype, shape in input_props]
    dtypes, shapes = zip(*input_props)
    queue = tf.PaddingFIFOQueue(capacity=10, dtypes=dtypes, shapes=shapes)
    self.enqueue_op = queue.enqueue(self.queue_input_tensors)
    self.input_tensors = queue.dequeue()

    self.predictions, self.loss = self.get_predictions_and_loss(*self.input_tensors)
    # bert stuff
    tvars = tf.trainable_variables()
    assignment_map, initialized_variable_names = modeling.get_assignment_map_from_checkpoint(tvars, config['init_checkpoint'])
    tf.train.init_from_checkpoint(config['init_checkpoint'], assignment_map)
    print("**** Trainable Variables ****")
    for var in tvars:
      init_string = ""
      if var.name in initialized_variable_names:
        init_string = ", *INIT_FROM_CKPT*"
      # tf.logging.info("  name = %s, shape = %s%s", var.name, var.shape,
                      # init_string)
      print("  name = %s, shape = %s%s", var.name, var.shape,
                      init_string)

    num_train_steps = int(
                    self.config['num_docs'] * self.config['num_epochs'])
    num_warmup_steps = int(num_train_steps * 0.1)
    self.global_step = tf.train.get_or_create_global_step()
    self.train_op = optimization.create_custom_optimizer(tvars,
                      self.loss, self.config['bert_learning_rate'], self.config['task_learning_rate'],
                      num_train_steps, num_warmup_steps, False, self.global_step, freeze=-1) 

# 需要导入模块: from bert import modeling [as 别名]
# 或者: from bert.modeling import get_assignment_map_from_checkpoint [as 别名]
def __init__(self, config):
    self.config = config
    self.max_segment_len = config['max_segment_len']
    self.max_span_width = config["max_span_width"]
    self.genres = { g:i for i,g in enumerate(config["genres"]) }
    self.subtoken_maps = {}
    self.gold = {}
    self.eval_data = None # Load eval data lazily.
    self.bert_config = modeling.BertConfig.from_json_file(config["bert_config_file"])
    self.tokenizer = tokenization.FullTokenizer(
                vocab_file=config['vocab_file'], do_lower_case=False)

    input_props = []
    input_props.append((tf.int32, [None, None])) # input_ids.
    input_props.append((tf.int32, [None, None])) # input_mask
    input_props.append((tf.int32, [None])) # Text lengths.
    input_props.append((tf.int32, [None, None])) # Speaker IDs.
    input_props.append((tf.int32, [])) # Genre.
    input_props.append((tf.bool, [])) # Is training.
    input_props.append((tf.int32, [None])) # Gold starts.
    input_props.append((tf.int32, [None])) # Gold ends.
    input_props.append((tf.int32, [None])) # Cluster ids.
    input_props.append((tf.int32, [None])) # Sentence Map

    self.queue_input_tensors = [tf.placeholder(dtype, shape) for dtype, shape in input_props]
    dtypes, shapes = zip(*input_props)
    queue = tf.PaddingFIFOQueue(capacity=10, dtypes=dtypes, shapes=shapes)
    self.enqueue_op = queue.enqueue(self.queue_input_tensors)
    self.input_tensors = queue.dequeue()

    self.predictions, self.loss = self.get_predictions_and_loss(*self.input_tensors)
    # bert stuff
    tvars = tf.trainable_variables()
    # If you're using TF weights only, tf_checkpoint and init_checkpoint can be the same
    # Get the assignment map from the tensorflow checkpoint. Depending on the extension, use TF/Pytorch to load weights.
    assignment_map, initialized_variable_names = modeling.get_assignment_map_from_checkpoint(tvars, config['tf_checkpoint'])
    init_from_checkpoint = tf.train.init_from_checkpoint if config['init_checkpoint'].endswith('ckpt') else load_from_pytorch_checkpoint
    init_from_checkpoint(config['init_checkpoint'], assignment_map)
    print("**** Trainable Variables ****")
    for var in tvars:
      init_string = ""
      if var.name in initialized_variable_names:
        init_string = ", *INIT_FROM_CKPT*"
      # tf.logging.info("  name = %s, shape = %s%s", var.name, var.shape,
                      # init_string)
      print("  name = %s, shape = %s%s" % (var.name, var.shape, init_string))

    num_train_steps = int(
                    self.config['num_docs'] * self.config['num_epochs'])
    num_warmup_steps = int(num_train_steps * 0.1)
    self.global_step = tf.train.get_or_create_global_step()
    self.train_op = optimization.create_custom_optimizer(tvars,
                      self.loss, self.config['bert_learning_rate'], self.config['task_learning_rate'],
                      num_train_steps, num_warmup_steps, False, self.global_step, freeze=-1,
                      task_opt=self.config['task_optimizer'], eps=config['adam_eps']) 

# 需要导入模块: from bert import modeling [as 别名]
# 或者: from bert.modeling import get_assignment_map_from_checkpoint [as 别名]
def __init__(self, config):
    self.config = config
    self.subtoken_maps = {}
    self.max_segment_len = config['max_segment_len']
    self.max_span_width = config["max_span_width"]
    self.genres = { g:i for i,g in enumerate(config["genres"]) }
    self.eval_data = None # Load eval data lazily.
    self.bert_config = modeling.BertConfig.from_json_file(config["bert_config_file"])
    self.sep = 102
    self.cls = 101
    self.tokenizer = tokenization.FullTokenizer(
                vocab_file=config['vocab_file'], do_lower_case=False)

    input_props = []
    input_props.append((tf.int32, [None, None])) # input_ids.
    input_props.append((tf.int32, [None, None])) # input_mask
    input_props.append((tf.int32, [None, None])) # input_ids.
    input_props.append((tf.int32, [None, None])) # input_mask
    input_props.append((tf.int32, [None])) # Text lengths.
    input_props.append((tf.int32, [None, None])) # Speaker IDs.
    input_props.append((tf.int32, [])) # Genre.
    input_props.append((tf.bool, [])) # Is training.
    input_props.append((tf.int32, [None])) # Gold starts.
    input_props.append((tf.int32, [None])) # Gold ends.
    input_props.append((tf.int32, [None])) # Cluster ids.
    input_props.append((tf.int32, [None])) # Sentence Map

    self.queue_input_tensors = [tf.placeholder(dtype, shape) for dtype, shape in input_props]
    dtypes, shapes = zip(*input_props)
    queue = tf.PaddingFIFOQueue(capacity=10, dtypes=dtypes, shapes=shapes)
    self.enqueue_op = queue.enqueue(self.queue_input_tensors)
    self.input_tensors = queue.dequeue()

    self.predictions, self.loss = self.get_predictions_and_loss(*self.input_tensors)
    # bert stuff
    tvars = tf.trainable_variables()
    assignment_map, initialized_variable_names = modeling.get_assignment_map_from_checkpoint(tvars, config['tf_checkpoint'])
    init_from_checkpoint = tf.train.init_from_checkpoint if config['init_checkpoint'].endswith('ckpt') else load_from_pytorch_checkpoint
    init_from_checkpoint(config['init_checkpoint'], assignment_map)
    print("**** Trainable Variables ****")
    for var in tvars:
      init_string = ""
      if var.name in initialized_variable_names:
        init_string = ", *INIT_FROM_CKPT*"
      # tf.logging.info("  name = %s, shape = %s%s", var.name, var.shape,
                      # init_string)
      print("  name = %s, shape = %s%s" % (var.name, var.shape, init_string))

    num_train_steps = int(
                    self.config['num_docs'] * self.config['num_epochs'])
    num_warmup_steps = int(num_train_steps * 0.1)
    self.global_step = tf.train.get_or_create_global_step()
    self.train_op = optimization.create_custom_optimizer(tvars,
                      self.loss, self.config['bert_learning_rate'], self.config['task_learning_rate'],
                      num_train_steps, num_warmup_steps, False, self.global_step, freeze=-1)