mindspore/model_zoo/official/nlp/bert/task_squad_config.yaml

112 lines
3.5 KiB
YAML

# Builtin Configurations(DO NOT CHANGE THESE CONFIGURATIONS unless you know exactly what you are doing)
enable_modelarts: False
# Url for modelarts
data_url: ""
train_url: ""
checkpoint_url: ""
# Path for local
data_path: "/cache/data"
output_path: "/cache/train"
load_path: "/cache/checkpoint_path"
device_target: "Ascend"
enable_profiling: False
# ==============================================================================
description: "run_squad"
do_train: "false"
do_eval: "false"
device_id: 0
epoch_num: 3
num_class: 2
train_data_shuffle: "true"
eval_data_shuffle: "false"
train_batch_size: 32
eval_batch_size: 1
vocab_file_path: ""
eval_json_path: ""
save_finetune_checkpoint_path: "./squad_finetune/ckpt/"
load_pretrain_checkpoint_path: ""
load_finetune_checkpoint_path: ""
train_data_file_path: ""
schema_file_path: ""
# export related
export_batch_size: 1
export_ckpt_file: ''
export_file_name: 'bert_squad'
file_format: 'AIR'
optimizer_cfg:
optimizer: 'Momentum'
AdamWeightDecay:
learning_rate: 0.0001 # 1e-4
end_learning_rate: 0.00000000001 # 1e-11
power: 5.0
weight_decay: 0.001 # 1e-3
decay_filter: ['layernorm', 'bias']
eps: 0.000001 # 1e-6
Lamb:
learning_rate: 0.0001 # 1e-4,
end_learning_rate: 0.00000000001 # 1e-11
power: 5.0
weight_decay: 0.01
decay_filter: ['layernorm', 'bias']
Momentum:
learning_rate: 0.0001 # 1e-4
momentum: 0.9
bert_net_cfg:
seq_length: 384
vocab_size: 30522
hidden_size: 768
num_hidden_layers: 12
num_attention_heads: 12
intermediate_size: 3072
hidden_act: "gelu"
hidden_dropout_prob: 0.1
attention_probs_dropout_prob: 0.1
max_position_embeddings: 512
type_vocab_size: 2
initializer_range: 0.02
use_relative_positions: False
dtype: mstype.float32
compute_type: mstype.float16
---
# Help description for each configuration
enable_modelarts: "Whether training on modelarts, default: False"
data_url: "Url for modelarts"
train_url: "Url for modelarts"
data_path: "The location of the input data."
output_path: "The location of the output file."
device_target: "Running platform, choose from Ascend or CPU, and default is Ascend."
enable_profiling: 'Whether enable profiling while training, default: False'
do_train: "Eable train, default is false"
do_eval: "Eable eval, default is false"
device_id: "Device id, default is 0."
epoch_num: "Epoch number, default is 1."
num_class: "The number of class, default is 2."
train_data_shuffle: "Enable train data shuffle, default is true"
eval_data_shuffle: "Enable eval data shuffle, default is false"
train_batch_size: "Train batch size, default is 32"
eval_batch_size: "Eval batch size, default is 1"
vocab_file_path: "Vocab file path"
eval_json_path: "Evaluation json file path, can be eval.json"
save_finetune_checkpoint_path: "Save checkpoint path"
load_pretrain_checkpoint_path: "Load checkpoint file path"
load_finetune_checkpoint_path: "Load checkpoint file path"
train_data_file_path: "Data path, it is better to use absolute path"
schema_file_path: "Schema path, it is better to use absolute path"
export_batch_size: "export batch size."
export_ckpt_file: "Bert ckpt file."
export_file_name: "bert output air name."
file_format: "file format"
---
# chocies
device_target: ['Ascend', 'GPU']
do_train: ["true", "false"]
do_eval: ["true", "false"]
train_data_shuffle: ["true", "false"]
eval_data_shuffle: ["true", "false"]
file_format: ["AIR", "ONNX", "MINDIR"]