Failed to load relative path images for multimodality models

[shaowei-su]

Reminder

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

System Info

[2025-03-13 01:39:21,513] [INFO] [real_accelerator.py:219:get_accelerator] Setting ds_accelerator to cuda (auto detect)

- `llamafactory` version: 0.9.1
- Platform: Linux-5.15.149-99.162.amzn2.x86_64-x86_64-with-glibc2.35
- Python version: 3.11.10
- PyTorch version: 2.4.1.post302 (GPU)
- Transformers version: 4.46.1
- Datasets version: 3.1.0
- Accelerate version: 1.0.1
- PEFT version: 0.12.0
- TRL version: 0.9.6
- GPU type: NVIDIA A100-SXM4-80GB
- DeepSpeed version: 0.15.4
- Bitsandbytes version: 0.44.1

Reproduction

Run example (https://github.com/hiyouga/LLaMA-Factory/blob/v0.9.2/examples/train_lora/qwen2vl_lora_sft.yaml) where the training arg `dataset_dir` is absolute path not under current directory.

Sample error logs:

  File "/mnt/conda-volume//lib/python3.11/site-packages/datasets/arrow_dataset.py", line 3073, in map
    for rank, done, content in Dataset._map_single(**dataset_kwargs):
  File "/mnt/conda-volume//lib/python3.11/site-packages/datasets/arrow_dataset.py", line 3476, in _map_single
    batch = apply_function_on_filtered_inputs(
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/conda-volume//lib/python3.11/site-packages/datasets/arrow_dataset.py", line 3338, in apply_function_on_filtered_inputs
    processed_inputs = function(*fn_args, *additional_args, **fn_kwargs)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/conda-volume//lib/python3.11/site-packages/llamafactory/data/processor/supervised.py", line 99, in preprocess_dataset
    input_ids, labels = self._encode_data_example(
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/conda-volume//lib/python3.11/site-packages/llamafactory/data/processor/supervised.py", line 43, in _encode_data_example
    messages = self.template.mm_plugin.process_messages(prompt + response, images, videos, audios, self.processor)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/conda-volume//lib/python3.11/site-packages/llamafactory/data/mm_plugin.py", line 1131, in process_messages
    mm_inputs = self._get_mm_inputs(images, videos, audios, processor)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/conda-volume//lib/python3.11/site-packages/llamafactory/data/mm_plugin.py", line 1095, in _get_mm_inputs
    images = self._regularize_images(
             ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/conda-volume//lib/python3.11/site-packages/llamafactory/data/mm_plugin.py", line 161, in _regularize_images
    image = Image.open(image)
            ^^^^^^^^^^^^^^^^^
  File "/mnt/conda-volume//lib/python3.11/site-packages/PIL/Image.py", line 3465, in open
    fp = builtins.open(filename, "rb")
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
FileNotFoundError: [Errno 2] No such file or directory: 'mllm_demo_data/1.jpg'

Others

No response