数据加载出现问题：pad_sequence(): argument 'padding_value' (position 3) must be float, not NoneType

[yanxp]

提交前必须检查以下项目

• [X] 请确保使用的是仓库最新代码（git pull），一些问题已被解决和修复。
• [X] 我已阅读项目文档和FAQ章节并且已在Issue中对问题进行了搜索，没有找到相似问题和解决方案。
• [X] 第三方插件问题：例如llama.cpp、LangChain、text-generation-webui等，同时建议到对应的项目中查找解决方案。

问题类型

其他问题

基础模型

Chinese-Alpaca-2 (7B/13B)

操作系统

Linux

详细描述问题

lr=1e-4
lora_rank=64
lora_alpha=128
lora_trainable="q_proj,v_proj,k_proj,o_proj,gate_proj,down_proj,up_proj"
modules_to_save="embed_tokens,lm_head"
lora_dropout=0.05

pretrained_model=/mnt/wx_feature/home/chopinyan/codework/Chinese-LLaMA-Alpaca-2/scripts/chinese-alpaca-2-7b/
chinese_tokenizer_path=/mnt/wx_feature/home/chopinyan/codework/Chinese-LLaMA-Alpaca-2/scripts/chinese-alpaca-2-7b/
dataset_dir=/mnt/wx_feature/home/chopinyan/codework/Chinese-LLaMA-Alpaca-2/scripts/firefly/
per_device_train_batch_size=1
per_device_eval_batch_size=1
gradient_accumulation_steps=1
max_seq_length=512
output_dir=output_dir
validation_file=/mnt/wx_feature/home/chopinyan/codework/Chinese-LLaMA-Alpaca-2/scripts/firefly/alpaca_data-0-3252.json

deepspeed_config_file=ds_zero2_no_offload.json

torchrun --nnodes 1 --nproc_per_node 1 run_clm_sft_with_peft.py \
    --deepspeed ${deepspeed_config_file} \
    --model_name_or_path ${pretrained_model} \
    --tokenizer_name_or_path ${chinese_tokenizer_path} \
    --dataset_dir ${dataset_dir} \
    --per_device_train_batch_size ${per_device_train_batch_size} \
    --per_device_eval_batch_size ${per_device_eval_batch_size} \
    --do_train \    --do_eval \
    --seed $RANDOM \
    --fp16 \
    --num_train_epochs 1 \
    --lr_scheduler_type cosine \
    --learning_rate ${lr} \
    --warmup_ratio 0.03 \
    --weight_decay 0 \
    --logging_strategy steps \
    --logging_steps 10 \
    --save_strategy steps \
    --save_total_limit 3 \
    --evaluation_strategy steps \
    --eval_steps 100 \
    --save_steps 200 \
    --gradient_accumulation_steps ${gradient_accumulation_steps} \
    --preprocessing_num_workers 8 \
    --max_seq_length ${max_seq_length} \
    --output_dir ${output_dir} \
    --overwrite_output_dir \
    --ddp_timeout 30000 \
    --logging_first_step True \
    --lora_rank ${lora_rank} \
    --lora_alpha ${lora_alpha} \
    --trainable ${lora_trainable} \
    --lora_dropout ${lora_dropout} \
    --modules_to_save ${modules_to_save} \
    --torch_dtype float16 \
    --validation_file ${validation_file} \
    --load_in_kbits 16 \
    --gradient_checkpointing \
    --ddp_find_unused_parameters False

依赖情况（代码类问题务必提供）

正常依赖

运行日志或截图

[INFO|trainer.py:386] 2023-11-13 20:05:21,910 >> You have loaded a model on multiple GPUs. `is_model_parallel` attribu
te will be force-set to `True` to avoid any unexpected behavior such as device placement mismatching.
Traceback (most recent call last):
  File "run_clm_sft_with_peft.py", line 531, in <module>
    main()
  File "run_clm_sft_with_peft.py", line 504, in main
    train_result = trainer.train(resume_from_checkpoint=checkpoint)
  File "/usr/local/python/lib/python3.8/site-packages/transformers/trainer.py", line 1544, in train
    return inner_training_loop(
  File "/usr/local/python/lib/python3.8/site-packages/transformers/trainer.py", line 1558, in _inner_training_loop
    train_dataloader = self.get_train_dataloader()
  File "/usr/local/python/lib/python3.8/site-packages/transformers/trainer.py", line 856, in get_train_dataloader
    for step, inputs in enumerate(train_dataloader):
  File "/usr/local/python/lib/python3.8/site-packages/accelerate/data_loader.py", line 451, in __iter__
    current_batch = next(dataloader_iter)
  File "/usr/local/python/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 634, in __next__
    data = self._next_data()
  File "/usr/local/python/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 678, in _next_data
    data = self._dataset_fetcher.fetch(index)  # may raise StopIteration
  File "/usr/local/python/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py", line 54, in fetch
    return self.collate_fn(data)
  File "/mnt/wx_feature/home/chopinyan/codework/Chinese-LLaMA-Alpaca-2/scripts/training/build_dataset.py", line 96, in
 __call__
    input_ids = torch.nn.utils.rnn.pad_sequence(
  File "/usr/local/python/lib/python3.8/site-packages/torch/nn/utils/rnn.py", line 399, in pad_sequence
    return torch._C._nn.pad_sequence(sequences, batch_first, padding_value)
TypeError: pad_sequence(): argument 'padding_value' (position 3) must be float, not NoneType
ERROR:torch.distributed.elastic.multiprocessing.api:failed (exitcode: 1) local_rank: 0 (pid: 129986) of binary: /usr/l
ocal/python/bin/python3.8

[iMountTai]

像是cache生成有问题，删掉cache重新生成一下试试

[yanxp]

像是cache生成有问题，删掉cache重新生成一下试试

删了重试，还是这个错误

[iMountTai]

不是很确定问题所在，代码在生成cache的过程中可能因为内存不足而程序失败的问题，重提就能解决，但是生成cache后训练中出现问题还没遇到

[iMountTai]

你实际输入的tokenizer不是我们发布的tokenizer，然后你的tokenizer.model中没有pad_token这个选项，所以会出现这个错误。

[yusufcakmakk]

I think we can add the following code block to sft trainer.

DEFAULT_PAD_TOKEN = "<pad>"
if tokenizer.pad_token is None:
   print(f"Adding pad token {DEFAULT_PAD_TOKEN}")
   tokenizer.add_special_tokens(dict(pad_token=DEFAULT_PAD_TOKEN))

[yusufcakmakk]

I have created pr for this issue.

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your consideration.

[github-actions[bot]]

Closing the issue, since no updates observed. Feel free to re-open if you need any further assistance.

[Shajiu]

你实际输入的tokenizer不是我们发布的tokenizer，然后你的tokenizer.model中没有pad_token这个选项，所以会出现这个错误。

自己训练词汇表时，怎么加入这个pad_token？

[Shajiu]

你实际输入的tokenizer不是我们发布的tokenizer，然后你的tokenizer.model中没有pad_token这个选项，所以会出现这个错误。

我的打印后有如下：

Generate config GenerationConfig {
  "bos_token_id": 1,
  "do_sample": true,
  "eos_token_id": 2,
  "max_length": 4096,
  "pad_token_id": 0,
  "temperature": 0.6,
  "top_p": 0.9
}

但是也出现问题了；