LitGPT fine-tuning Dont Use GPU

[strikene]

This is the first time I have used LitGPT for fine-tuning, and as an experiment, I chose Qwen2.5-0.5B for fine-tuning.
I am Runing this on windows server 2022dc

(.env) f:\litgpt>litgpt finetune Qwen/Qwen2.5-0.5B-Instruct --data JSON --data.json_path F:\litgpt\alpaca_zh_demo.json --data.val_split_fraction 0.1 --out_dir f:\out
I found that the whole process was done using the CPU, not my GPU

I confirm that cuda12.2 is installed correctly and I can use ML Studio to use the GGUF format model.

Did I miss any settings?