with torch.no_grad():
    base_model_output = model(**token_dict)

transformer_embeddings = base_model_output.last_hidden_state
transformer_embeddings