### model
model_name_or_path: Qwen/Qwen2.5-1.5B
trust_remote_code: true

### method
stage: sft
do_train: true
finetuning_type: lora
lora_rank: 64
lora_target: all

### dataset
dataset: QAtrain
eval_dataset: QAval
template: qwen
cutoff_len: 2048
# max_samples: 50
overwrite_cache: true
preprocessing_num_workers: 8

### output
resume_from_checkpoint: /kaggle/working/Model/last-checkpoint
output_dir: /kaggle/working/
logging_steps: 50
save_steps: 200
plot_loss: true
overwrite_output_dir: true

### train
per_device_train_batch_size: 1
gradient_accumulation_steps: 16
learning_rate: 1.0e-4
num_train_epochs: 3.0
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: true # full
ddp_timeout: 180000000
torch_compile: false
### eval
# val_size: 0.1
per_device_eval_batch_size: 1
eval_strategy: steps
eval_steps: 200

report_to: wandb
run_name: Qwennn

push_to_hub: true
export_hub_model_id: Youssef/QWEN_Arabic_Q&A
hub_private_repo: true
hub_strategy: checkpoint