I’ve been unable to push models to the hub during training. Using the following training arguments I receive the messages below:
args = TrainingArguments(
output_dir=“OnlyPhishGPT2_50K”,
overwrite_output_dir = True,
per_device_train_batch_size=128,
per_device_eval_batch_size=128,
evaluation_strategy=“epoch”,
logging_strategy=“epoch”,
gradient_accumulation_steps=1,
num_train_epochs=10,
weight_decay=0.1,
warmup_steps=1_000,
lr_scheduler_type=“cosine”,
learning_rate=5e-4,
save_strategy= “epoch”,
save_total_limit = 1,
fp16=True,
load_best_model_at_end = True,
metric_for_best_model = “loss”,
push_to_hub=True,
disable_tqdm = False
)
On the hub, the output directory is created but after training it only contains the .gitattributes file.
I can still push datasets and models to the hub.
I’ve updated to Transformers 4.27 and I have git-lfs installed.