TypeError: sequence item 139: expected str instance, NoneType found

I used verl 0.5.0.dev with vllm==0.10.1 to train the model Intern-S1-mini, but got the error `TypeError: sequence item 139: expected str instance, NoneType found` during decoding.

The error lines were
"python3.10/site-packages/transformers/tokenization_utils.py", line 1114, in _decode
string = self.convert_tokens_to_string(current_sub_text)
"InternS1/tokenization_interns1.py", line 899, in convert_tokens_to_string
text = "".join(tokens)

So I tried to print what token id caused such `NoneType` token, finding the token ids 153058, 153162, and 153055, even larger than the vocabulary size 152971.

Does anyone know what the problem is?

By the way, the logging gives the warnings (maybe for vllms),
`Only support config type of {'llama', 'gemma3_text', 'qwen2', 'minicpmo', 'qwen2_vl', 'mistral', 'qwen2_moe', 'minicpmv', 'qwen2_5_vl', 'qwen3_moe', 'qwen3', 'deepseek_v3'}, but got interns1. MFU will always be zero.`
Maybe the problems come from here?