423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-14 10:56:56 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix initializing data arguments

Browse Source 
Former-commit-id: 18f87c1b25
This commit is contained in:
hiyouga
2023-06-27 22:50:23 +08:00
parent 307f5866e9
commit e877b8d55b
3 changed files with 35 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
src/utils/common.py
View File

@@ -103,11 +103,10 @@ def _init_adapter(
lastest_checkpoint = None
if model_args.checkpoint_dir is not None:
if not os.path.exists(os.path.join(model_args.checkpoint_dir[0], WEIGHTS_NAME)):
raise ValueError("Provided path ({}) does not contain a LoRA weight.".format(model_args.checkpoint_dir[0]))
if not os.path.exists(os.path.join(model_args.checkpoint_dir[0], CONFIG_NAME)):
raise ValueError("The given checkpoint may be not a LoRA checkpoint, \
please specify `--finetuning_type full/freeze` instead.")
assert os.path.exists(os.path.join(model_args.checkpoint_dir[0], WEIGHTS_NAME)), \
"Provided path ({}) does not contain a LoRA weight.".format(model_args.checkpoint_dir[0])
assert os.path.exists(os.path.join(model_args.checkpoint_dir[0], CONFIG_NAME)), \
"The given checkpoint may be not a LoRA checkpoint, please specify `--finetuning_type full/freeze` instead."
if (is_trainable and model_args.resume_lora_training) or (not is_mergeable): # continually train on the lora weights
checkpoints_to_merge, lastest_checkpoint = model_args.checkpoint_dir[:-1], model_args.checkpoint_dir[-1]
@@ -267,6 +266,8 @@ def prepare_args(
transformers.utils.logging.enable_explicit_format()
# Check arguments (do not check finetuning_args since it may be loaded from checkpoints)
data_args.init_for_training()
if stage != "sft" and training_args.predict_with_generate:
raise ValueError("`predict_with_generate` cannot be set as True at PT, RM and PPO stages.")

4
src/utils/config.py
View File

@@ -134,7 +134,7 @@ class DataTrainingArguments:
)
source_prefix: Optional[str] = field(
default=None,
metadata={"help": "A prefix to add before every source text. Use `|` to separate multiple prefixes."}
metadata={"help": "A prefix to add before every source text. Use `|` to separate multiple prefixes in training."}
)
dev_ratio: Optional[float] = field(
default=0,
@@ -145,7 +145,7 @@ class DataTrainingArguments:
metadata={"help": "Which template to use for constructing prompts in training and inference."}
)
def __post_init__(self): # support mixing multiple datasets
def init_for_training(self): # support mixing multiple datasets
dataset_names = [ds.strip() for ds in self.dataset.split(",")]
with open(os.path.join(self.dataset_dir, "dataset_info.json"), "r") as f:
dataset_info = json.load(f)