### model
model_name_or_path: model_weights/qwen3-4b-base

### method
stage: sft
do_train: true
finetuning_type: full
deepspeed: examples/deepspeed/ds_z0_config.json 
enable_liger_kernel: true
packing: true


### dataset
dataset: openthoughts3_300k_qwen3-8b
template: qwen3
cutoff_len: 16384
overwrite_cache: true
preprocessing_num_workers: 16
dataloader_persistent_workers: true
dataloader_pin_memory: true
dataloader_num_workers: 4

### output
logging_steps: 1
save_steps: 100
save_total_limit: 10
plot_loss: true
overwrite_output_dir: false
save_only_model: false
#report_to: wandb
report_to: none
run_name: qwen3-4b-base-open-thoughts3-qwen3-8b

### train
per_device_train_batch_size: 4
gradient_accumulation_steps: 2
learning_rate: 0.00008
max_steps: 3000
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: true
ddp_timeout: 180000000