Untitled

import pandas as pd
data = pd.read_parquet('/home/t12/QuangNV/mini_project_stp/notebooks/res/0000.parquet', engine='pyarrow')

with open('/home/t12/QuangNV/mini_project_stp/data/raw/train/train.en', 'a') as file:
    for i in range(len(data)):
        file.write(data['translation'][i]['en']+'\n')

with open('/home/t12/QuangNV/mini_project_stp/data/raw/train/train.vi', 'a') as file:
    for i in range(len(data)):
        file.write(data['translation'][i]['vi']+'\n')

data = pd.read_parquet('/home/t12/QuangNV/mini_project_stp/notebooks/res/iwslt2015-en-vi_validation_0000.parquet', engine='pyarrow')

with open('/home/t12/QuangNV/mini_project_stp/data/raw/val/val.en', 'a') as file:
    for i in range(len(data)):
        file.write(data['translation'][i]['en']+'\n')

with open('/home/t12/QuangNV/mini_project_stp/data/raw/val/val.vi', 'a') as file:
    for i in range(len(data)):
        file.write(data['translation'][i]['vi']+'\n')

data = pd.read_parquet('/home/t12/QuangNV/mini_project_stp/notebooks/res/iwslt2015-en-vi_test_0000.parquet', engine='pyarrow')

with open('/home/t12/QuangNV/mini_project_stp/data/raw/test/test.en', 'a') as file:
    for i in range(len(data)):
        file.write(data['translation'][i]['en']+'\n')

with open('/home/t12/QuangNV/mini_project_stp/data/raw/test/test.vi', 'a') as file:
    for i in range(len(data)):
        file.write(data['translation'][i]['vi']+'\n')

RAW='/home/t12/QuangNV/mini_project_stp/data/raw'
TMP='/home/t12/QuangNV/mini_project_stp/data/tmp'
DUMP='/home/t12/QuangNV/mini_project_stp/data/dump'

!python3 /home/t12/QuangNV/mini_project_stp/src/opennmt/scripts/bert_tokenize.py \
    --bert bert-base-multilingual-cased \
    --prefixes $RAW/train/train.en $RAW/train/train.vi $RAW/val/val.en $RAW/val/val.vi $RAW/test/test.en $RAW/test/test.vi \
    --output_dir $TMP

!python3 /home/t12/QuangNV/mini_project_stp/src/opennmt/scripts/bert_prepro.py --src $TMP/train.vi.bert \
                                   --tgt $TMP/train.en.bert \
                                   --output $DUMP/VIEN.db

VSIZE=200000
FREQ=0
SHARD_SIZE=200000
!python3 /home/t12/QuangNV/mini_project_stp/src/opennmt/preprocess.py \
    -train_src $TMP/train.vi.bert \
    -train_tgt $TMP/train.en.bert \
    -valid_src $TMP/val.vi.bert \
    -valid_tgt $TMP/val.en.bert \
    -save_data $DUMP/VIEN \
    -src_seq_length 150 \
    -tgt_seq_length 150 \
    -src_vocab_size $VSIZE \
    -tgt_vocab_size $VSIZE \
    -vocab_size_multiple 8 \
    -src_words_min_frequency $FREQ \
    -tgt_words_min_frequency $FREQ \
    -share_vocab \
    -shard_size $SHARD_SIZE

!mv $TMP/val.en.bert $DUMP/dev.en.bert
!mv $TMP/test.en.bert $DUMP/test.en.bert
!mv $TMP/val.vi.bert $DUMP/dev.vi.bert
!mv $TMP/test.vi.bert $DUMP/test.vi.bert
!cp /home/t12/QuangNV/mini_project_stp/data/raw/val/val.en /home/t12/QuangNV/mini_project_stp/data/dump/ref/dev.en
!cp /home/t12/QuangNV/mini_project_stp/data/raw/test/test.en /home/t12/QuangNV/mini_project_stp/data/dump/ref/test.en

!python3 /home/t12/QuangNV/mini_project_stp/src/run_cmlm_finetuning.py --train_file /home/t12/QuangNV/mini_project_stp/data/dump/VIEN.db \
                                    --vocab_file /home/t12/QuangNV/mini_project_stp/data/dump/VIEN.vocab.pt \
                                    --valid_src /home/t12/QuangNV/mini_project_stp/data/dump/dev.vi.bert \
                                    --valid_tgt /home/t12/QuangNV/mini_project_stp/data/dump/dev.en.bert \
                                    --bert_model bert-base-multilingual-cased \
                                    --output_dir /home/t12/QuangNV/mini_project_stp/output/out \
                                    --train_batch_size 2048 \
                                    --learning_rate 5e-5 \
                                    --valid_steps 5000 \
                                    --num_train_steps 41000 \
                                    --warmup_proportion 0.05 \
                                    --gradient_accumulation_steps 1 \
                                    # --fp16