clip-indonesian / hybrid_clip /run_training.sh
Galuh Sahid
Add code
bb13925
#!/bin/bash
SCRIPT_DIR=.
MODEL_DIR=/mnt/disks/data-1/models/training_v4_unfreeze
IMAGE_ENCODER="openai/clip-vit-base-patch32"
TEXT_ENCODER="flax-community/indonesian-roberta-base"
python ${SCRIPT_DIR}/run_hybrid_clip.py \
--output_dir ${MODEL_DIR} \
--overwrite_output_dir \
--tokenizer_name=${TEXT_ENCODER} \
--train_file="../data/train_dataset_v6.json" \
--validation_file="../data/val_dataset_v6.json" \
--do_train --do_eval \
--num_train_epochs="20" --max_seq_length 96 \
--per_device_train_batch_size="64" \
--per_device_eval_batch_size="64" \
--learning_rate="0.00001" --warmup_ratio 0.1 --weight_decay 0.0 \
--preprocessing_num_workers 16 \
--exp_name training_v4_unfreeze \
--text_model_name_or_path=${TEXT_ENCODER} \
--vision_model_name_or_path=${IMAGE_ENCODER} \
--eval_steps 500 \
--logging_steps 100 \
--save_steps 500 \
--save_total_limit 5 \
--log_wandb \
--run_from_checkpoint="/mnt/disks/data-1/models/training_v4/ckpt-70999" # edit
#--freeze_backbones
#--push_to_hub