bangla-gpt2 / run_training.sh
Ritobrata Ghosh
Saving weights and logs of epoch 1
da170c3
raw
history blame contribute delete
637 Bytes
./run_clm_flax.py \
--output_dir="${MODEL_DIR}" \
--model_type="gpt2" \
--config_name="${MODEL_DIR}" \
--tokenizer_name="${MODEL_DIR}" \
--dataset_name="mc4" \
--dataset_config_name="bn" \
--do_train --do_eval \
--block_size="256" \
--per_device_train_batch_size="32" \
--per_device_eval_batch_size="32" \
--learning_rate="5e-3" --warmup_steps="1000" \
--adam_beta1="0.9" --adam_beta2="0.98" --weight_decay="0.01" \
--overwrite_output_dir \
--num_train_epochs="3" \
--logging_steps="50" \
--save_steps="2500" \
--eval_steps="2500" \
--push_to_hub