How to get logs for every x number of steps instead of epochs

Right now we are getting logs for each 0.5 epoch but I want logs for each x number of steps. how i can achieve this

Description:
I am currently using the training script and noticed that the logs are generated every 0.25 epochs. For my use-case, it would be more beneficial to get logs every x number of steps. This granularity would help in better monitoring and understanding of the training dynamics over shorter intervals.

Details:

Current Behavior: Logs are generated every 0.25 epochs.
Desired Behavior: Option to log every 10 number of steps.

Example of current logs:
{'loss': 2.8187, 'learning_rate': 2e-05, 'epoch': 0.25}
{'loss': 2.843, 'learning_rate': 1.959492736144978e-05, 'epoch': 0.5}
...

Use-case:
When training models, especially with larger datasets, epochs can take a significant amount of time. Having logs for every x number of steps would offer more frequent insights into the model's performance and learning rate adjustments.

Please help and suggest code to fix this issue.

Current command

torchrun --nproc_per_node=8 --master_port=20001 fastchat/train/train_mem.py \
  --model_name_or_path meta-llama/Llama-2-7b-hf  \
  --data_path data/dummy_conversation.json \
  --bf16 True \
  --output_dir output_vicuna_13b \
  --num_train_epochs 3 \
  --per_device_train_batch_size 4 \
  --per_device_eval_batch_size 32 \
  --gradient_accumulation_steps 4 \
  --evaluation_strategy "steps" \
  --eval_steps 1500 \
  --save_strategy "steps" \
  --save_steps 1500 \
  --save_total_limit 8 \
  --learning_rate 2e-5 \
  --weight_decay 0. \
  --warmup_ratio 0.04 \
  --lr_scheduler_type "cosine" \
  --logging_steps 1 \
  --fsdp "full_shard auto_wrap offload" \
  --fsdp_transformer_layer_cls_to_wrap 'LlamaDecoderLayer' \
  --tf32 True \
  --model_max_length 2048 \
  --gradient_checkpointing True \
  --lazy_preprocess False

Start with fastchat/train/train_mem.py and trace the --logging_steps argument in the provided torchrun command. Confirm whether the existing training configuration supports step-based logging; done means logs are emitted at the requested interval, such as every 10 steps, rather than only at epoch intervals.