llava-phi/scripts/llava_phi/get_base_model.sh

#!/bin/bash

python llava_phi/train/convert_model2base_llava_phi.py \
    --model_name_or_path susnato/phi-2 \
    --version plain \
    --data_path pretrain_data/blip_sample.json \
    --image_folder pretrain_data/blip_images \
    --vision_tower ./clip-vit-large-patch14-336 \
    --mm_projector_type mlp2x_gelu \
    --tune_mm_mlp_adapter True \
    --mm_vision_select_layer -2 \
    --mm_use_im_start_end False \
    --mm_use_im_patch_token False \
    --bf16 True \
    --output_dir ./base_checkpoints_llava_phi \
    --num_train_epochs 1 \
    --per_device_train_batch_size 4 \
    --per_device_eval_batch_size 4 \
    --gradient_accumulation_steps 2 \
    --evaluation_strategy "no" \
    --save_strategy "steps" \
    --save_steps 24000 \
    --save_total_limit 1 \
    --learning_rate 1e-3 \
    --weight_decay 0.1 \
    --warmup_ratio 0. \
    --lr_scheduler_type "cosine" \
    --logging_steps 10 \
    --tf32 False \
    --model_max_length 2048 \
    --gradient_checkpointing True \
    --dataloader_num_workers 4 \
    --lazy_preprocess True \
    --report_to tensorboard