llama3_lora_sft_glm4.yaml

# model
model_name_or_path: THUDM/glm-4-9b

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

# ddp
ddp_timeout: 180000000

# dataset
dataset: alpaca_gpt4_en,alpaca_gpt4_zh,ICLR2017,ICLR2018,ICLR2019,ICLR2020,ICLR2021,ICLR2022,ICLR2023
template: glm4
cutoff_len: 1024
max_samples: 1000
val_size: 0.1
overwrite_cache: true
preprocessing_num_workers: 16

# output
output_dir: saves/glm-4-9b(iclr)/lora/sft
logging_steps: 10
save_steps: 500
plot_loss: true
overwrite_output_dir: true

# train
per_device_train_batch_size: 1
gradient_accumulation_steps: 2
learning_rate: 0.0001
num_train_epochs: 10.0
lr_scheduler_type: cosine
warmup_steps: 0.1
fp16: true

# eval
per_device_eval_batch_size: 1
evaluation_strategy: steps
eval_steps: 500

rope_scaling: dynamic