| #!/usr/bin/env bash |
| python3 run_clm_flax.py \ |
| --output_dir="./" \ |
| --model_type="gpt2" \ |
| --config_name="./" \ |
| --tokenizer_name="./" \ |
| --dataset_name="oscar" \ |
| --dataset_config_name="unshuffled_deduplicated_th" \ |
| --do_train \ |
| --do_eval \ |
| --block_size="512" \ |
| --preprocessing_num_workers="64" \ |
| --weight_decay="0.01" \ |
| --per_device_train_batch_size="64" \ |
| --per_device_eval_batch_size="64" \ |
| --learning_rate="5e-3" \ |
| --warmup_steps="1000" \ |
| --overwrite_output_dir \ |
| --num_train_epochs="3" \ |
| --adam_beta1="0.9" \ |
| --adam_beta2="0.999" \ |
| --adam_epsilon="1e-8" \ |
| --push_to_hub |
|
|
