Inputs and input_ids includes labels??! 输入token包含全部输出这对吗？

[Luffy966]

Reminder

• I have read the above rules and searched the existing issues.

System Info

• llamafactory version: 0.9.3.dev0
• Platform: Linux-5.4.0-162-generic-x86_64-with-glibc2.31
• Python version: 3.12.9
• PyTorch version: 2.6.0+cu118 (GPU)
• Transformers version: 4.49.0
• Datasets version: 3.4.1
• Accelerate version: 1.5.2
• PEFT version: 0.15.0
• TRL version: 0.9.6
• GPU type: NVIDIA GeForce RTX 4090
• GPU number: 8
• GPU memory: 23.65GB
• DeepSpeed version: 0.16.4
• vLLM version: 0.8.1

Reproduction

I found by chance in logs that in the printed training example, the inputs tokens include labels tokens, which is very abnormal!!

I first feel like it's my fault, then I start to randomly read some issues that show training example. And I also found this happens in every dataset or example. The following log is from: #7156 (comment)

training example:
input_ids:
[151644, 8948, 198, 2610, 525, 264, 10950, 17847, 13, 151645, 198, 151644, 872, 198, 151652, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151656, 151653, 74785, 279, 1887, 14878, 323, 862, 19026, 311, 279, 2766, 13, 151645, 198, 151644, 77091, 198, 32, 5220, 448, 2805, 6869, 11, 10078, 8664, 11, 12233, 264, 4158, 1909, 323, 3691, 24549, 14638, 304, 4065, 315, 279, 1965, 11, 12771, 705, 264, 2311, 504, 279, 1965, 11, 323, 8930, 432, 311, 1349, 151645, 198]
inputs:
<|im_start|>system
You are a helpful assistant.<|im_end|>
<|im_start|>user
<|vision_start|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|video_pad|><|vision_end|>Describe the main subjects and their contributions to the video.<|im_end|>
<|im_start|>assistant
A woman with short hair, slightly fat, wearing a white top and black pants stood in front of the table, picked up a book from the table, and opened it to read<|im_end|>

label_ids:
[-100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, 32, 5220, 448, 2805, 6869, 11, 10078, 8664, 11, 12233, 264, 4158, 1909, 323, 3691, 24549, 14638, 304, 4065, 315, 279, 1965, 11, 12771, 705, 264, 2311, 504, 279, 1965, 11, 323, 8930, 432, 311, 1349, 151645, 198]
labels:
A woman with short hair, slightly fat, wearing a white top and black pants stood in front of the table, picked up a book from the table, and opened it to read<|im_end|>

I refer to the source code in llamafactory and found these several lines in SupervisedDatasetProcessor._encode_data_example() from src/llamafactory/data/processor/supervised.py are suspected:

if self.data_args.mask_history:  # reversed sequences
    input_ids = source_ids + target_ids + input_ids
    labels = source_label + target_label + labels
else:
    input_ids += source_ids + target_ids
    labels += source_label + target_label

No matter mask_history is turned on or not, target_ids will always be added into input_ids. Is it normal? @hiyouga

Others

No response

[Luffy966]

Since it's an AutoRegressive task, inputs should include labels.