#!/bin/bash

export HF_PROJECT="t5-base-dutch"

# Variables for training the tokenizer and creating the config
export VOCAB_SIZE="32000"
export N_INPUT_SENTENCES="1000000" # Num of sentences to train the tokenizer
export DATASET="yhavinga/mc4_nl_cleaned" # Name of the dataset in the Huggingface Hub
export DATASET_CONFIG="full" # Config of the dataset in the Huggingface Hub
export DATASET_SPLIT="train" # Split to use for training tokenizer and model
export TEXT_FIELD="text" # Field containing the text to be used for training
export CONFIG_TYPE="t5-base" # Config that our model will use
export MODEL_PATH="${HOME}/data/${HF_PROJECT}" # Path to the model, e.g. here inside the mount

python run_t5_mlm_flax.py \
    --output_dir="${MODEL_PATH}" \
    --model_type="t5" \
    --config_name="${MODEL_PATH}" \
    --tokenizer_name="${MODEL_PATH}" \
    --preprocessing_num_workers="96" \
    --do_train --do_eval \
    --dataset_name="${DATASET}" \
    --dataset_config_name="${DATASET_CONFIG}" \
    --max_seq_length="512" \
    --per_device_train_batch_size="16" \
    --per_device_eval_batch_size="16" \
    --adafactor \
    --learning_rate="0.005" \
    --overwrite_output_dir \
    --num_train_epochs="1" \
    --logging_steps="500" \
    --save_steps="80000" \
    --eval_steps="2500" \
    --weight_decay="0.01" \
    --warmup_steps="10000" \
    --validation_split_count="15000" \
    --push_to_hub