bf16: true
cutoff_len: 1024
dataset: my_chartQA
dataset_dir: data
ddp_timeout: 180000000
do_train: true
eval_steps: 500
eval_strategy: steps
finetuning_type: lora
flash_attn: auto
gradient_accumulation_steps: 8
include_num_input_tokens_seen: true
learning_rate: 1.0e-05
logging_steps: 50
lora_alpha: 16
lora_dropout: 0
lora_rank: 8
lora_target: all
lr_scheduler_type: cosine
max_grad_norm: 1.0
max_samples: 100000
model_name_or_path: google/paligemma-3b-pt-448
num_train_epochs: 2.0
optim: adamw_hf
output_dir: saves/PaliGemma-3B-pt-448-Chat/lora/train_2024-10-15-09-54-06
packing: false
per_device_eval_batch_size: 2
per_device_train_batch_size: 2
plot_loss: true
preprocessing_num_workers: 16
report_to: none
save_steps: 500
stage: sft
template: paligemma
val_size: 0.05
warmup_steps: 1