Qwen

/

Qwen2-VL-7B-Instruct

like 602

Incorporating generation tag to chat_template.json so that return_assistant_tokens_mask can work correctly with the tokenizer (https://github.com/huggingface/transformers/pull/30650/files)

@yangapku @chenkq Would very much appreciate it if maintainors can take a look and merge this at your earliest convenience. Having this generation tag makes it much easier to obtain assistant token masks which helps with constructing training labels when finetuning.