推理遇到数据集加载问题

[BinBrent]

Reminder

• I have read the README and searched the existing issues.

Reproduction

显示无法读取数据文件，但命令行中已经输出了数据样本

Running tokenizer on dataset (num_proc=8): 100%|██████████| 50/50 [00:00<00:00, 119.29 examples/s]
Running tokenizer on dataset (num_proc=8): 100%|██████████| 50/50 [00:00<00:00, 118.64 examples/s]
multiprocess.pool.RemoteTraceback:
"""
Traceback (most recent call last):
File "/home//.conda/envs/LLaMA/lib/python3.10/site-packages/multiprocess/pool.py", line 125, in worker
result = (True, func(args, kwds))
File "/home//.conda/envs/LLaMA/lib/python3.10/site-packages/datasets/utils/py_utils.py", line 678, in _write_generator_to_queue
for i, result in enumerate(func(kwargs)):
File "/home//.conda/envs/LLaMA/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 3595, in _map_single
os.chmod(cache_file_name, 0o666 & ~umask)
FileNotFoundError: [Errno 2] No such file or directory: '/home/*/.cache/huggingface/datasets/json/default-83535c0c955e8ee5/0.0.0/c8d2d9508a2a2067ab02cd118834ecef34c3700d143b31835ec4235bf10109f7/cache-6939a289e8425e94_00000_of_00008.arrow'
"""

Expected behavior

input_ids:
[21586, 63, 684, 9226, 100, 14482, 100, 4440, 45, 28595, 731, 303, 1547, 303, 14991, 341, 10984, 7692, 451, 331, 1168, 451, 36822, 3966, 51, 465, 10485, 63, 303, 29113, 327, 687, 731, 418, 1168, 451, 29113, 451, 36822, 3966, 51, 465, 3777, 63, 303, 1916, 63, 906, 10984, 7692, 451, 341, 36822, 3966, 51, 303, 1547, 222, 35222, 63, 244]
inputs:
Human: def calculate_average_price(prices):
"""
Calculate the average price of a list of fashion items.

Args:
prices (list): A list of prices of fashion items.

Returns:
float: The average price of the fashion items.
"""

Assistant:

System Info

transformers 4.40.2
torch 2.2.0+cu11.8
deepspeed 0.14.2
datasets 2.19.1
accelerate 0.30.1

Others

No response

[BinBrent]

使用了full_multi_gpu里的llama3_full_predict.sh

 model
#model_name_or_path: /home/***/models/Meta-Llama-3-70B-Instruct
model_name_or_path: /***/models/starcoder2-15b-instruct-v0.1
# method
stage: sft
do_predict: true
finetuning_type: full

# dataset
dataset: inference
template: default
cutoff_len: 4096
max_samples: 50
overwrite_cache: true
preprocessing_num_workers: 8

# output
output_dir: saves/starcoder-instruct/predict
overwrite_output_dir: true

#fp16: true
# eval
per_device_eval_batch_size: 1
#auto_find_batch_size: true

[hiyouga]

删除 cache 文件夹