model_id: ./models/final_ft_2_epochs_lr1e-05_llava-phi_full
model_family: llava-phi
LoRA:
  r: 0
  alpha: 128
  dropout: 0.05
loss_type: grad_ascent
tune_vision_tower: false
tune_mm_projector: true
tune_language_model: true
data_path: ./dataset/full.json
split: full
batch_size: 8
gradient_accumulation_steps: 4
max_grad_norm: 1.0
num_epochs: 2
save_dir: models/final_ft_${num_epochs}_epochs_lr${lr}_${model_family}_${split}
save_steps: 180
lr: 1.0e-05
weight_decay: 0
seed: 233
workers: 4
lr_scheduler_type: cosine
warmup_ratio: 0
max_train_steps: -1
report_to: wandb
resume_from_checkpoint: ''