uvos/QRotaryTraining
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

remove stray breakpoint, fix mypy warning

Browse source
This commit is contained in:
Carl Philipp Klemm 2024-05-07 15:05:22 +02:00
parent 68f748e99e
commit ce2ada2617
2 changed files with 1 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

2
arguments.py
View file

@ -92,5 +92,5 @@ class TrainingArguments():
max_instant_params: int = field(default=0, metadata={"help": "Maximum amount of paramters to optimize per step in millions"}) max_instant_params: int = field(default=0, metadata={"help": "Maximum amount of paramters to optimize per step in millions"})
churn_percent: int = field(default=100, metadata={"help": "The percentage of active parameters to replace when changeing active parameters"}) churn_percent: int = field(default=100, metadata={"help": "The percentage of active parameters to replace when changeing active parameters"})
eval_steps: int = field(default=-1, metadata={"help": "Number of optimization steps after wich to compute the evaluation loss"}) eval_steps: int = field(default=-1, metadata={"help": "Number of optimization steps after wich to compute the evaluation loss"})
eval_prompt: str = field(default=None, metadata={"help": "A prompt to used during eval to check if the model is learning"}) eval_prompt: str | None = field(default=None, metadata={"help": "A prompt to used during eval to check if the model is learning"})
reshufle_steps: int = field(default=50, metadata={"help": "Number of steps to take before changing the active parameters"}) reshufle_steps: int = field(default=50, metadata={"help": "Number of steps to take before changing the active parameters"})

1
datamodules.py
View file

@ -218,7 +218,6 @@ def create_data_module(tokenizer: transformers.PreTrainedTokenizer, data_args: D
eval_dataset = dataset['eval'] eval_dataset = dataset['eval']
else: else:
print('Splitting train dataset in train and validation according to `eval_dataset_size`') print('Splitting train dataset in train and validation according to `eval_dataset_size`')
breakpoint()
dataset = dataset['train'].train_test_split( dataset = dataset['train'].train_test_split(
test_size=data_args.eval_dataset_size, shuffle=True, seed=42 test_size=data_args.eval_dataset_size, shuffle=True, seed=42
) )