bf16: true
cutoff_len: 8192
dataset: train_split2
dataset_dir: data
ddp_timeout: 180000000
do_train: true
double_quantization: true
enable_thinking: true
finetuning_type: lora
flash_attn: auto
gradient_accumulation_steps: 8
include_num_input_tokens_seen: true
label_smoothing_factor: 0.0
learning_rate: 5.0e-05
logging_steps: 10
lora_alpha: 64
lora_dropout: 0
lora_rank: 32
lora_target: all
lr_scheduler_type: cosine
max_grad_norm: 1.0
max_samples: 3716
model_name_or_path: zai-org/GLM-Z1-9B-0414
num_train_epochs: 3.0
optim: adamw_8bit
output_dir: saves/GLM-Z1-0414-9B-Chat/lora/glm-treinado
packing: false
per_device_train_batch_size: 4
plot_loss: true
preprocessing_num_workers: 16
quantization_bit: 4
quantization_method: bnb
report_to: none
save_steps: 1000
stage: sft
template: alpaca
trust_remote_code: true
use_rslora: true
use_unsloth: true
warmup_steps: 500
weight_decay: 0.1