cutoff_len: 2048
dataset: farcaster_pretrain
dataset_dir: data
ddp_timeout: 180000000
do_train: true
finetuning_type: lora
flash_attn: auto
fp16: true
gradient_accumulation_steps: 8
include_num_input_tokens_seen: true
learning_rate: 0.0002
logging_steps: 1
lora_alpha: 16
lora_dropout: 0
lora_rank: 256
lora_target: all
lr_scheduler_type: cosine
max_grad_norm: 1.0
max_samples: 100000
model_name_or_path: Qwen/Qwen2-7B-Instruct
num_train_epochs: 8.0
optim: adamw_8bit
output_dir: saves/Qwen2-7B-Chat/lora/final-pretrain-v2
packing: true
per_device_train_batch_size: 1
plot_loss: true
preprocessing_num_workers: 16
report_to: none
save_steps: 100
stage: pt
template: qwen
warmup_steps: 0