423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-17 12:20:37 +08:00
Code Issues Packages Projects Releases Wiki Activity

implement efficient packing without cross-contamination attention

Browse Source 
Former-commit-id: b2c367bc61
This commit is contained in:
ancv
2024-06-12 11:56:01 +07:00
parent 1a261add61
commit 045eb155a2
9 changed files with 287 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/llamafactory/hparams/data_args.py
View File

@@ -84,6 +84,10 @@ class DataArguments:
"help": "Whether or not to pack the sequences in training. Will automatically enable in pre-training."
},
)
efficient_packing: Optional[bool] = field(
default=None,
metadata={"help": "Whether or not to pack the sequences without cross-contamination attention for efficient training."},
)
tokenized_path: Optional[str] = field(
default=None,
metadata={"help": "Path to save or load the tokenized datasets."},