How to Load a pre-trained model from disk with Huggingface Transformers

To load a pre-trained model from a disk using the Hugging Face Transformers library, save the pre-trained model and its tokenizer to your local disk, and then you can load them using the from_pretrained.

Follow the below step-by-step guide.

1. Install the Hugging Face Transformers library using this command if you haven’t already.

   pip install transformers

2. Save the pre-trained model and its tokenizer to your local disk.

   from transformers import BertModel, BertTokenizer
   
   # Choose a pre-trained model, e.g., bert-base-uncased
   model_name = "bert-base-uncased"
   
   # Download the pre-trained model and tokenizer
   model = BertModel.from_pretrained(model_name)
   tokenizer = BertTokenizer.from_pretrained(model_name)
   
   # Save the model and tokenizer to your local disk
   model.save_pretrained("path/to/your/local/directory/model")
   tokenizer.save_pretrained("path/to/your/local/directory/tokenizer")

   Replace these two paths “path/to/your/local/directory/model” and “path/to/your/local/directory/tokenizer” with your desired directory paths.

3. Load the pre-trained model and tokenizer from your local disk.

   from transformers import BertModel, BertTokenizer
   
   # Load the model and tokenizer from the local disk
   model = BertModel.from_pretrained("path/to/your/local/directory/model")
   tokenizer = BertTokenizer.from_pretrained("path/to/your/local/directory/tokenizer")

   Again, replace the “path/to/your/local/directory/model” and “path/to/your/local/directory/tokenizer” with the directory paths where you saved the model and tokenizer.

That’s it.