infrastructure-dependency-version-c-when-loading-pytorch-model-weights-with-transforme-48b39016

THEN Convert your model checkpoint files from PyTorch .bin format to safetensors format. Load the checkpoint with an older transformers version (e.g., 4.41.0) and resave using `model.save_pretrained(save_directory, safe_serialization=True)`, or use a machine with torch>=2.6 (e.g., Google Colab) to perform the conversion. Safetensors do not have the security risk and are the default recommended format.

# Load with older transformers, then resave as safetensors
model = AutoModel.from_pretrained('path/to/bin/checkpoint')
model.save_pretrained('new/path', safe_serialization=True)