[BUG] AttributeError: can't set attribute when training chatglm2-6b

[wangshuai09]

Describe the bug
Training chatglm2-6b reported a AttributeError.
Traceback (most recent call last): File "fastchat/train/train.py", line 304, in <module> train() File "fastchat/train/train.py", line 283, in train tokenizer.pad_token = tokenizer.unk_token AttributeError: can't set attribute

To Reproduce
Model download form huggingface-chatglm2-6b
Script is
torchrun --nproc_per_node=4 --master_port=20001 fastchat/train/train.py \ --model_name_or_path /home/xxx/models/chatglm2-6b \ --data_path /home/xxx/datasets/evol-instruct-chinese/evol-instruct-chinese-1024-subset.json \ --fp16 True \ --output_dir output_chatglm \ --num_train_epochs 5 \ --per_device_train_batch_size 8 \ --per_device_eval_batch_size 1 \ --gradient_accumulation_steps 1 \ --evaluation_strategy "no" \ --save_strategy "epoch" \ --learning_rate 5e-5 \ --weight_decay 0. \ --lr_scheduler_type "cosine" \ --logging_steps 1 \ --fsdp "full_shard auto_wrap" \ --model_max_length 512 \ --gradient_checkpointing True \ --lazy_preprocess True

Reason
In chatglm2-6b code, pad_token is readonly