seed: 42
run_mode: 'train'
output_dir: './output' # path to save checkpoint/strategy
load_checkpoint: ''
src_strategy_path_or_dir: ''
auto_trans_ckpt: False  # If true, auto transform load_checkpoint to load in distributed model
only_save_strategy: False
resume_training: False

# context
context:
  mode: 0 #0--Graph Mode; 1--Pynative Mode
  device_target: "Ascend"
  enable_graph_kernel: False
  max_call_depth: 10000
  save_graphs: False
  device_id: 0

# aicc
remote_save_url: "Please input obs url on AICC platform."

# runner
runner_config:
  epochs: 3
  batch_size: 24
  sink_mode: False
  sink_size: 2
runner_wrapper:
  type: TrainOneStepCell


# parallel
use_parallel: False
parallel:
  parallel_mode: 0 # 0-data parallel, 1-semi-auto parallel, 2-auto parallel, 3-hybrid parallel
  gradients_mean: True
  enable_alltoall: False
  full_batch: False
  search_mode: "sharding_propagation"
  enable_parallel_optimizer: False
  strategy_ckpt_save_file: "./ckpt_strategy.ckpt"
parallel_config:
  data_parallel: 1
  model_parallel: 1
  expert_parallel: 1
  pipeline_stage: 1
  micro_batch_num: 1
  gradient_aggregation_group: 4
micro_batch_interleave_num: 1

# moe
moe_config:
  expert_num: 1
  capacity_factor: 1.05
  aux_loss_factor: 0.05
  num_experts_chosen: 1

# recompute
recompute: False
parallel_optimizer_comm_recompute: False
mp_comm_recompute: True
recompute_slice_activation: False

# autotune
auto_tune: False
filepath_prefix: './autotune'
autotune_per_step: 10

# profile
profile: False
profile_start_step: 1
profile_stop_step: 10
init_start_profile: False
profile_communication: False
profile_memory: True

# Trainer
trainer:
  type: TokenClassificationTrainer
  model_name: tokcls_bert_base_chinese_cluener
# if True, do evaluate during the training process. if false, do nothing.
# note that the task trainer should support _evaluate_in_training function.
do_eval: False

# train dataset
train_dataset: &train_dataset
  data_loader:
    type: CLUENERDataLoader
    dataset_dir: "./cluener/"
    stage: "train"
    column_names: ["text", "label_id"]
  text_transforms:
    type: TokenizeWithLabel
    max_length: 128
    padding: "max_length"
  label_transforms:
    type: LabelPadding
    max_length: 128
    padding_value: 0
  tokenizer:
    type: BertTokenizer
    cls_token: '[CLS]'
    mask_token: '[MASK]'
    pad_token: '[PAD]'
    sep_token: '[SEP]'
    unk_token: '[UNK]'
    is_tokenize_char: True
    do_lower_case: False
    checkpoint_name_or_path: tokcls_bert_base_chinese
  input_columns: ["text", "label_id"]
  output_columns: ["input_ids", "token_type_ids", "attention_mask", "label_id"]
  column_order: ["input_ids", "token_type_ids", "attention_mask", "label_id"]
  num_parallel_workers: 8
  python_multiprocessing: False
  drop_remainder: True
  batch_size: 24
  repeat: 1
  numa_enable: False
  prefetch_size: 30
  seed: 2022
train_dataset_task:
  type: TokenClassificationDataset
  dataset_config: *train_dataset

# eval dataset
eval_dataset: &eval_dataset
  data_loader:
    type: CLUENERDataLoader
    dataset_dir: "./cluener/"
    stage: "dev"
    column_names: ["text", "label_id"]
  text_transforms:
    type: TokenizeWithLabel
    max_length: 128
    padding: "max_length"
  label_transforms:
    type: LabelPadding
    max_length: 128
    padding_value: 0
  tokenizer:
    type: BertTokenizer
    cls_token: '[CLS]'
    mask_token: '[MASK]'
    pad_token: '[PAD]'
    sep_token: '[SEP]'
    unk_token: '[UNK]'
    is_tokenize_char: True
    do_lower_case: False
    checkpoint_name_or_path: tokcls_bert_base_chinese
  input_columns: ["text", "label_id"]
  output_columns: ["input_ids", "token_type_ids", "attention_mask", "label_id"]
  column_order: ["input_ids", "token_type_ids", "attention_mask", "label_id"]
  num_parallel_workers: 8
  python_multiprocessing: False
  drop_remainder: True
  batch_size: 24
  repeat: 1
  numa_enable: False
  prefetch_size: 30
  seed: 2022
eval_dataset_task:
  type: TokenClassificationDataset
  dataset_config: *eval_dataset

# model
model:
  model_config:
    type: BertConfig
    use_one_hot_embeddings: False
    num_labels: 31
    dropout_prob: 0.1
    batch_size: 24
    seq_length: 128 # length of input sentence
    vocab_size: 21128 # size of vocab
    embedding_size: 768 # size of text feature
    num_layers: 12 # model depth
    num_heads: 12 # number of attention heads
    expand_ratio: 4
    hidden_act: "gelu" # activation
    post_layernorm_residual: True # select postlayernorm or prelayernorm
    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
    use_past: False
    compute_dtype: "float32"
    checkpoint_name_or_path: "tokcls_bert_base_chinese_cluener"
  arch:
    type: BertForTokenClassification

# lr schedule
lr_schedule:
  type: linear
  learning_rate: 0.00003  # 3e-5
  warmup_ratio: 0.1
  total_steps: -1 # -1 means it will load the total steps of the dataset
layer_scale: False
layer_decay: 0.65

# optimizer
optimizer:
  type: adamw
  weight_decay: 0.01
  eps: 0.00000001  # 1e-8
lr_scale: False
lr_scale_factor: 256

# callbacks
callbacks:
  - type: MFLossMonitor
  - type: CheckpointMonitor
    prefix: "mindformers"
    save_checkpoint_steps: 100
    integrated_save: True
    async_save: False
eval_callbacks:
  - type: ObsMonitor

# metric
metric:
  type: EntityScore

# processor
processor:
  type: BertProcessor
  return_tensors: ms
  tokenizer:
    type: BertTokenizer
    cls_token: '[CLS]'
    mask_token: '[MASK]'
    pad_token: '[PAD]'
    sep_token: '[SEP]'
    unk_token: '[UNK]'
    is_tokenize_char: True
    do_lower_case: False
    checkpoint_name_or_path: tokcls_bert_base_chinese