38 lines
1.4 KiB
Bash
Executable File
38 lines
1.4 KiB
Bash
Executable File
#!/bin/bash
|
|
|
|
export HF_PROJECT="gpt2-large-dutch-2"
|
|
|
|
# Variables for training the tokenizer and creating the config
|
|
export VOCAB_SIZE="50257"
|
|
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="gpt2-large" # 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_clm_flax.py \
|
|
--output_dir="${MODEL_PATH}" \
|
|
--model_type="gpt2" \
|
|
--config_name="${MODEL_PATH}" \
|
|
--model_name_or_path="${MODEL_PATH}" \
|
|
--tokenizer_name="${MODEL_PATH}" \
|
|
--preprocessing_num_workers="96" \
|
|
--do_train --do_eval \
|
|
--dataset_name="${DATASET}" \
|
|
--dataset_config_name="${DATASET_CONFIG}" \
|
|
--block_size="512" \
|
|
--per_device_train_batch_size="4" \
|
|
--per_device_eval_batch_size="4" \
|
|
--learning_rate="0.000033" --warmup_steps="5000" \
|
|
--adafactor \
|
|
--overwrite_output_dir \
|
|
--num_train_epochs="1" \
|
|
--logging_steps="500" \
|
|
--save_steps="20000" \
|
|
--eval_steps="2500"
|
|
|
|
# \
|
|
# --push_to_hub
|
|
# --adam_beta1="0.9" --adam_beta2="0.98" --weight_decay="0.01" \
|