File size: 1,519 Bytes
be0e2ec | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 | # Config CHỈ task commitment (trích từ config/train.yml gốc, bỏ env/soc/gov/specificity).
# Tune: python -m src.training.tune_hyperparams --config config/train_commitment.yml --task commitment --trials 10
# Train: python -m src.training.train_model --config config/train_commitment.yml --task commitment
defaults:
task: commitment
_model_subst: &model_subst
name: vinai/phobert-base-v2
max_length: 256
word_segment: true
_tune_subst: &tune_subst
train_batch_size: [8, 16]
_data: &data
use_context_prev: false
use_context_next: false
_es: &es
enabled: true
patience: 2
threshold: 0.0
_train: &train
epochs: 10
seeds: [42, 43, 44, 45, 46]
train_batch_size: 16
eval_batch_size: 32
learning_rate: 2.0e-5
weight_decay: 0.01
warmup_ratio: 0.1
use_class_weights: true
weight_method: sqrt_inverse
gradient_accumulation_steps: 1
max_grad_norm: 1.0
label_smoothing_factor: 0.0
precision: bf16
metric_for_best_model: macro_f1
tasks:
commitment:
labels: [0, 1]
seed: 42
model: *model_subst
data: *data
tune: *tune_subst
paths:
train_data: data/vi_gold/commitment/train.parquet
val_data: data/vi_gold/commitment/val.parquet
test_data: data/vi_gold/commitment/test.parquet
output_dir: outputs/models/commitment
training: *train
early_stopping: *es
|