No checkpoints saved with full finetune even when I set --save_steps

[cauwulixuan]

I tried to finetune model with the following scripts:

bash scripts/run_finetune.sh

As I can see, the commands in this script was like:

#!/bin/bash
# Please run this script under ${project_id} in project directory of
#   https://github.com/shizhediao/llm-ft
#     COMMIT: d5fecf30ba8011067b10cf51fede53a5ab6574e4

# Parses arguments
model_name_or_path=gpt2
dataset_path=data/alpaca/train
output_dir=output_models/finetune
deepspeed_args="--master_port=11000"

while [[ $# -ge 1 ]]; do
  key="$1"
  case ${key} in
    -m|--model_name_or_path)
      model_name_or_path="$2"
      shift
      ;;
    -d|--dataset_path)
      dataset_path="$2"
      shift
      ;;
    -o|--output_model_path)
      output_dir="$2"
      shift
      ;;
    --deepspeed_args)
      deepspeed_args="$2"
      shift
      ;;
    *)
      echo "error: unknown option \"${key}\"" 1>&2
      exit 1
  esac
  shift
done

# Finetune
exp_id=finetune
project_dir=$(cd "$(dirname $0)"/..; pwd)
log_dir=${project_dir}/log/${exp_id}
mkdir -p ${output_dir} ${log_dir}

deepspeed ${deepspeed_args} \
  examples/finetune.py \
    --model_name_or_path ${model_name_or_path} \
    --dataset_path ${dataset_path} \
    --output_dir ${output_dir} --overwrite_output_dir \
    --num_train_epochs 5 \
    --learning_rate 1e-4 \
    --block_size 512 \
    --per_device_train_batch_size 1 \
    --deepspeed configs/ds_config_zero3.json \
    --fp16 \
    --run_name finetune \
    --validation_split_percentage 0 \
    --logging_steps 20 \
    --do_train \
    --ddp_timeout 72000 \
    --save_steps 5000 \
    --dataloader_num_workers 1 \
    | tee ${log_dir}/train.log \
    2> ${log_dir}/train.err

After finetune, I got a final pytorch_model.bin in ${output_dir}, no other checkpints were saved. So I was confused, what does the arg --save_steps 5000 mean in this situation. Is it normal that no checkpoints-xxx saved after every 5000 steps ?

Any suggestions would be appriciate.

[research4pan]

Thanks for your interest in LMFlow! Normally --save_steps 5000 will save a checkpoint every 5000 steps. If it is not working, you may explicitly specify --save_strategy "steps". Hope that can be helpful 😄