config.ini

################################### 
#            configuration        #
###################################
[DEFAULT]
task_name = semeval
output_dir=/tmp/%(task_name)s
use_entity_indicator=True

[MODEL]
seed = 12345
#pretrained_model_name=bert-base-uncased
pretrained_model_name=bert-large-uncased
#pretrained_model_name=./large-uncased-model

[Train]
num_train_epochs=5.0
# Total number of training epochs to perform.
learning_rate=2e-5
# The initial learning rate for Adam.
per_gpu_train_batch_size=16
# Batch size per GPU/CPU for training.
per_gpu_eval_batch_size=8
# Batch size per GPU/CPU for evaluation.  
no_cuda=False
# Avoid using CUDA when available

[Dataset]
data_dir= ./data
 
max_seq_len=128

train=True
eval=True
evaluate_during_training=True

gradient_accumulation_steps=1 
# Number of updates steps to accumulate before performing a backward/update pass.
 
weight_decay=0.9
# Weight deay if we apply some.
adam_epsilon=1e-8
# Epsilon for Adam optimizer.
max_grad_norm=1.0
# Max gradient norm.

max_steps=-1
# If > 0: set total number of training steps to perform. Override num_train_epochs.
warmup_steps=0
# Linear warmup over warmup_steps.
logging_steps=200
# Log every X updates steps.

save_steps=200
# help="Save checkpoint every X updates steps.

eval_all_checkpoints=False
# Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number
l2_reg_lambda=5e-3
#l2_reg_lambda=0

overwrite_output_dir=True
# Overwrite the content of the output directory
overwrite_cache=True
# Overwrite the cached training and evaluation sets
local_rank=-1
# For distributed training: local_rank

latent_entity_typing=False