fix multimodal data and max_token arguments when using vllm tokenizer (text mode)#4

wangshangsam merged 7 commits intoCentML:mlperf-inf-mm-q3vl-v6.0CentML/dynamo:mlperf-inf-mm-q3vl-v6.0from

soodoshll:fix-mm-inputsoodoshll/dynamo:fix-mm-inputCopy head branch name to clipboard