amank
Made change to cleaning code, modified number of warmpu step, getting eval samples from validation split
7839b8e
| python3 -c "import jax; print(jax.devices())" | |
| ./run_mlm_flax_stream.py \ | |
| --output_dir="${MODEL_DIR}" \ | |
| --model_type="roberta" \ | |
| --config_name="${MODEL_DIR}" \ | |
| --tokenizer_name="${MODEL_DIR}" \ | |
| --dataset_name="mc4" \ | |
| --dataset_config_name="hi" \ | |
| --max_seq_length="256" \ | |
| --per_device_train_batch_size="128" \ | |
| --per_device_eval_batch_size="128" \ | |
| --learning_rate="3e-4" \ | |
| --warmup_steps="10000" \ | |
| --overwrite_output_dir \ | |
| --adam_beta1="0.9" \ | |
| --adam_beta2="0.98" \ | |
| --num_train_steps="10000" \ | |
| --num_eval_samples="5000" \ | |
| --preprocessing_num_workers="90" \ | |
| --logging_steps="250" \ | |
| --eval_steps="1000" |