IlyaGusev

{
    "max_tokens": 150,
    "min_agreement": 0.79,
    "epochs": 3,
    "eval_steps": 32,
    "warmup_steps": 16,
    "lr": 0.00003,
    "seed": 42,
    "batch_size": 16,
    "grad_accum_steps": 4,

IlyaGusev

import argparse
import json
import random

import torch
from torch.utils.data import Dataset
from transformers import AutoTokenizer, AutoModelForSequenceClassification
from transformers import Trainer, TrainingArguments
from tqdm import tqdm

IlyaGusev

import numpy as np
from tqdm import tqdm
import tensorflow as tf
import tensorflow_hub as hub
import tensorflow_text as text

DEFAULT_ENCODER_PATH = "https://tfhub.dev/google/LaBSE/2"
DEFAULT_PREPROCESSOR_PATH = "https://tfhub.dev/google/universal-sentence-encoder-cmlm/multilingual-preprocess/2"

IlyaGusev

rmq:
    image: rabbitmq:3.8.7
    container_name: rabbitmq
    volumes:
      - /data/rabbitmq/etc/:/etc/rabbitmq/
      - /data/rabbitmq/data/:/var/lib/rabbitmq/
      - /data/rabbitmq/logs/:/var/log/rabbitmq/
    ports:
      - 5672:5672
      - 15672:15672

IlyaGusev

import torch
from tqdm import tqdm
from transformers import BartForConditionalGeneration, MBartTokenizer

def chunks(lst, n):
    """Yield successive n-sized chunks from lst."""
    for i in range(0, len(lst), n):
        yield lst[i : i + n]

def generate_summaries(examples, model_name, batch_size):

IlyaGusev

import torch
from transformers import (
    BartConfig,
    BartForConditionalGeneration
)

def remove_ignore_keys_(state_dict):
    ignore_keys = [
        "encoder.version",
        "decoder.version",

IlyaGusev

BART_PATH="mbart.cc25"
TASK="data"
rm -rf "${TASK}-bin/"
fairseq-preprocess \
    --source-lang "source" \
    --target-lang "target" \
    --trainpref "${TASK}/train.bpe" \
    --validpref "${TASK}/val.bpe" \
    --testpref "${TASK}/test.bpe" \
    --destdir "${TASK}-bin/" \

IlyaGusev

import torch
from setuptools import setup, find_packages
import subprocess

import sys
import warnings
import os

# ninja build does not work unless include_dirs are abs path
this_dir = os.path.dirname(os.path.abspath(__file__))

IlyaGusev

FAIRSEQ_PATH="fairseq"
CHECKPOINT_PATH="checkpoints/checkpoint_best.pt"
DATA_BIN_PATH="data-bin"
BART_PATH="mbart.cc25"
langs=ar_AR,cs_CZ,de_DE,en_XX,es_XX,et_EE,fi_FI,fr_XX,gu_IN,hi_IN,it_IT,ja_XX,kk_KZ,ko_KR,lt_LT,lv_LV,my_MM,ne_NP,nl_XX,ro_RO,ru_RU,si_LK,tr_TR,vi_VN,zh_CN
python3.7 "${FAIRSEQ_PATH}"/generate.py "${DATA_BIN_PATH}" --path "${CHECKPOINT_PATH}" \
  --task translation_from_pretrained_bart --gen-subset test -t target -s source \
  --bpe 'sentencepiece' --sentencepiece-vocab "${BART_PATH}/sentence.bpe.model" \
  --sacrebleu --max-sentences 32 --langs $langs > predicted.txt

IlyaGusev

# Image: nvidia/cuda:11.7.0-cudnn8-devel-ubuntu20.04 

touch ~/.no_auto_tmux
apt-get update
apt-get install -y vim screen wget g++ git python3.8 python3.8-dev python3-distutils git-lfs
wget https://www.dropbox.com/s/cxws7jpt3nlxn2l/vimrc -O ~/.vimrc
wget https://www.dropbox.com/s/wbdlntxmujpat9o/screenrc -O ~/.screenrc
wget https://bootstrap.pypa.io/get-pip.py && python3.8 get-pip.py
python3.8 -m pip install torch transformers