423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-08-23 22:32:54 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #5909 from hiyouga/hiyouga/dev2

Browse Source 
[data] support auto convert for single image, add image_dir argument

Former-commit-id: bd08b8c441c47076faa03cc1efde21b22f14f058
This commit is contained in:
hoshi-hiyouga 2024-11-02 13:43:04 +08:00 committed by GitHub
commit 4b2c47fcae
2 changed files with 30 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
src/llamafactory/data/aligner.py
View File

@ -33,41 +33,47 @@ logger = get_logger(__name__)
def _convert_images( def _convert_images(
images: Sequence["ImageInput"], images: Union["ImageInput", Sequence["ImageInput"]],
dataset_attr: "DatasetAttr", dataset_attr: "DatasetAttr",
data_args: "DataArguments", data_args: "DataArguments",
) -> Optional[List["ImageInput"]]: ) -> Optional[List["ImageInput"]]:
r""" r"""
Optionally concatenates image path to dataset dir when loading from local disk. Optionally concatenates image path to dataset dir when loading from local disk.
""" """
if len(images) == 0: if not isinstance(images, list):
images = [images]
elif len(images) == 0:
return None return None
else:
images = images[:]
images = images[:]
if dataset_attr.load_from in ["script", "file"]: if dataset_attr.load_from in ["script", "file"]:
for i in range(len(images)): for i in range(len(images)):
if isinstance(images[i], str) and os.path.isfile(os.path.join(data_args.dataset_dir, images[i])): if isinstance(images[i], str) and os.path.isfile(os.path.join(data_args.image_dir, images[i])):
images[i] = os.path.join(data_args.dataset_dir, images[i]) images[i] = os.path.join(data_args.image_dir, images[i])
return images return images
def _convert_videos( def _convert_videos(
videos: Sequence["VideoInput"], videos: Union["VideoInput", Sequence["VideoInput"]],
dataset_attr: "DatasetAttr", dataset_attr: "DatasetAttr",
data_args: "DataArguments", data_args: "DataArguments",
) -> Optional[List["VideoInput"]]: ) -> Optional[List["VideoInput"]]:
r""" r"""
Optionally concatenates video path to dataset dir when loading from local disk. Optionally concatenates video path to dataset dir when loading from local disk.
""" """
if len(videos) == 0: if not isinstance(videos, list):
videos = [videos]
elif len(videos) == 0:
return None return None
else:
videos = videos[:]
videos = videos[:]
if dataset_attr.load_from in ["script", "file"]: if dataset_attr.load_from in ["script", "file"]:
for i in range(len(videos)): for i in range(len(videos)):
if isinstance(videos[i], str) and os.path.isfile(os.path.join(data_args.dataset_dir, videos[i])): if isinstance(videos[i], str) and os.path.isfile(os.path.join(data_args.image_dir, videos[i])):
videos[i] = os.path.join(data_args.dataset_dir, videos[i]) videos[i] = os.path.join(data_args.image_dir, videos[i])
return videos return videos

15
src/llamafactory/hparams/data_args.py
View File

@ -41,6 +41,10 @@ class DataArguments:
default="data", default="data",
metadata={"help": "Path to the folder containing the datasets."}, metadata={"help": "Path to the folder containing the datasets."},
) )
image_dir: Optional[str] = field(
default=None,
metadata={"help": "Path to the folder containing the images or videos. Defaults to `dataset_dir`."},
)
cutoff_len: int = field( cutoff_len: int = field(
default=1024, default=1024,
metadata={"help": "The cutoff length of the tokenized inputs in the dataset."}, metadata={"help": "The cutoff length of the tokenized inputs in the dataset."},
@ -111,7 +115,13 @@ class DataArguments:
) )
tokenized_path: Optional[str] = field( tokenized_path: Optional[str] = field(
default=None, default=None,
metadata={"help": "Path to save or load the tokenized datasets."}, metadata={
"help": (
"Path to save or load the tokenized datasets. "
"If tokenized_path not exists, it will save the tokenized datasets. "
"If tokenized_path exists, it will load the tokenized datasets."
)
},
) )
def __post_init__(self): def __post_init__(self):
@ -123,6 +133,9 @@ class DataArguments:
self.dataset = split_arg(self.dataset) self.dataset = split_arg(self.dataset)
self.eval_dataset = split_arg(self.eval_dataset) self.eval_dataset = split_arg(self.eval_dataset)
if self.image_dir is None:
self.image_dir = self.dataset_dir
if self.dataset is None and self.val_size > 1e-6: if self.dataset is None and self.val_size > 1e-6:
raise ValueError("Cannot specify `val_size` if `dataset` is None.") raise ValueError("Cannot specify `val_size` if `dataset` is None.")