max_seq_length = 512
per_device_train_batch_size = 2,
gradient_accumulation_steps = 4,
warmup_steps = 2,
max_steps = 10,
learning_rate = 0.0005,
fp16 = not torch.cuda.is_bf16_supported(),
bf16 = torch.cuda.is_bf16_supported(),
logging_steps = 1,
optim = "adamw_8bit",
weight_decay = 0.01,
lr_scheduler_type = "linear",
seed = 3407,
output_dir = "outputs",