Skip to content

LoRA微调embedding时,支持多模态数据混合训练吗,数据中包含单模态文到文图到图、跨模态图文以及混合多模态数据,目前直接将各个类型的数据写在一个jsonl里报错如下: #4895

Open
@viviansmlie

Description

@viviansmlie

[INFO:swift] Traceback (most recent call last):
File "/mnt/vv/ms-swift/swift/llm/dataset/utils.py", line 100, in getitem
return self.encode_func(data)
File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
return func(*args, **kwargs)
File "/mnt/vv/ms-swift/swift/llm/template/base.py", line 473, in encode
self._preprocess_inputs(inputs)
File "/mnt/vv/ms-swift/swift/llm/template/base.py", line 248, in _preprocess_inputs
images[i] = self._load_image(images[i], load_images)
File "/mnt/vv/ms-swift/swift/llm/template/base.py", line 162, in _load_image
image = load_image(image)
File "/mnt/vv/ms-swift/swift/llm/template/vision_utils.py", line 132, in load_image
image = load_file(image)
File "/mnt/vv/ms-swift/swift/llm/template/vision_utils.py", line 116, in load_file
with open(path, 'rb') as f:
IsADirectoryError: [Errno 21] Is a directory: '/mnt/vv/ms-swift'

训练数据的jsonl的格式是这样的:
{"query": "一个女人在剥土豆。", "response": "女人剥土豆。", "images": [""]} {"query": "喷泉喷水", "response": "<image>图像中有一个巨大的喷泉,形状xxxxxxxxxxxxxx。", "images": ["/mnt/images/000001/0000000001.jpg"]} {"query": "喷泉喷水", "response": "<image>, "images": ["/mnt/images/000001/0000000001.jpg"]} {"query": "<image>", "response": "<image>", "images": ["/mnt/images/000001/0000000001.jpg", "/mnt/target_images/000001/0000000001_pos.jpg"]} {"query": "<image>", "response": "图像中有一个巨大的喷泉,形状xxxxxxxxxxxxxx。", "images": ["/mnt/images/000001/0000000001.jpg"]}

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions