'NoneType' object has no attribute 'vocab_size', when I used nemorun cli to finetuning

[CaptainRui1000]

I was trying to fine-tune a qwen2_500m, so I used a command like pre-train.
nemorun llm finetune --factory qwen2_500m trainer.devices=1 trainer.strategy.context_parallel_size=1
But I got an error, AttributeError: 'NoneType' object has no attribute 'vocab_size'.

Then I tried to fine-tune Gemma2b, still the same error. I debugged the nemorun program, and found all recipe didn't set any tokenizer, right? So how can I make my experiment select any tokenizer with nemorun cli, is it possible?