03 Mar How to download a model from Hugging Face

Downloading models from Hugging Face is simple and can be done using the Transformers library or directly from the Hugging Face Hub. Below is a step-by-step guide to help you download and use models from Hugging Face.

Step 1: Install the Transformers Library

If you haven’t already installed the transformers library, you can do so using pip. On Google Colab, use the following command to install:

!pip install transformers

Step 2: Download a Model Using the Transformers Library

You can download a model using the from_pretrained method. This method downloads the model weights, configuration, and tokenizer (if applicable) from the Hugging Face Hub.

Example: Download a Pre-Trained BERT Model

from transformers import AutoModel, AutoTokenizer

# Download the model and tokenizer
model_name = "bert-base-uncased"
model = AutoModel.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Use the model and tokenizer
inputs = tokenizer("Hello, Hugging Face!", return_tensors="pt")
outputs = model(**inputs)
print(outputs.last_hidden_state.shape)  # Example output shape

Above, the output is:

torch.Size([1, 7, 768])

This shape is commonly seen in models like BERT, where each token in a sequence is represented by a 768-dimensional vector.

When we use the bert-base-uncased model and pass the input Hello, Hugging Face!, the last hidden state output shape represents the tensor dimensions. For this example, the shape you would typically see is:

torch.Size([1, 7, 768])

Here’s a breakdown:

• 1: The batch size—since there’s one input sentence.
• 7: The sequence length—this corresponds to the tokenized version of "Hello, Hugging Face!", including special tokens like [CLS] and [SEP].
• 768: The hidden size—each token is represented as a 768-dimensional vector, standard for BERT’s base architecture.

Step 3: Download a Model for a Specific Task

Hugging Face provides task-specific models (e.g., for text classification, question answering, etc.). You can download these models using the appropriate class.

Example: Download a Text Classification Model

from transformers import AutoModelForSequenceClassification, AutoTokenizer

# Download a model for text classification
model_name = "distilbert-base-uncased-finetuned-sst-2-english"
model = AutoModelForSequenceClassification.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Use the model for inference
inputs = tokenizer("I love Hugging Face!", return_tensors="pt")
outputs = model(**inputs)
print(outputs.logits)  # Output logits for classification

Step 4: Download a Model from the Hugging Face Hub Website

If you prefer to download models manually, you can do so from the Hugging Face Hub website:

1. Go to the Hugging Face Hub: https://huggingface.co/models.
2. Search for the model you want (e.g., bert-base-uncased).
3. Click on the model to open its page.
4. Download the model files directly from the “Files” tab.

Step 5: Use a Downloaded Model Locally

If you’ve downloaded a model manually, you can load it from a local directory:

from transformers import AutoModel, AutoTokenizer

# Load the model and tokenizer from a local directory
model = AutoModel.from_pretrained("path/to/local/model")
tokenizer = AutoTokenizer.from_pretrained("path/to/local/tokenizer")

Step 6: Download a Model with Custom Configurations

Some models have multiple configurations or variants. You can specify the configuration when downloading the model.

Example: Download a Multilingual BERT Model

from transformers import AutoModel, AutoTokenizer

# Download a multilingual BERT model
model_name = "bert-base-multilingual-cased"
model = AutoModel.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

Step 7: Download a Model with a Specific Framework

You can specify the framework (PyTorch, TensorFlow, or JAX) when downloading a model.

Example: Download a TensorFlow Model

from transformers import TFAutoModel, AutoTokenizer

# Download a TensorFlow model
model_name = "bert-base-uncased"
model = TFAutoModel.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

Step 8: Use the Model for Inference

Once the model is downloaded, you can use it for inference. Here’s an example of using a text classification model:

from transformers import pipeline

# Load a text classification pipeline
classifier = pipeline("text-classification", model="distilbert-base-uncased-finetuned-sst-2-english")

# Perform inference
result = classifier("I love Hugging Face!")
print(result)

Step 9: Save a Downloaded Model Locally

You can save a downloaded model and tokenizer to a local directory for future use:

# Save the model and tokenizer
model.save_pretrained("path/to/save/model")
tokenizer.save_pretrained("path/to/save/tokenizer")

To load the saved model later:

from transformers import AutoModel, AutoTokenizer

# Load the model and tokenizer from a local directory
model = AutoModel.from_pretrained("path/to/save/model")
tokenizer = AutoTokenizer.from_pretrained("path/to/save/tokenizer")

We used the following commands above:

If you liked the tutorial, spread the word and share the link and our website Studyopedia with others.

For Videos, Join Our YouTube Channel: Join Now

Read More:

• RAG Tutorial
• Generative AI Tutorial
• Machine Learning Tutorial
• Deep Learning Tutorial
• Ollama Tutorial
• Retrieval Augmented Generation (RAG) Tutorial
• Copilot Tutorial
• ChatGPT Tutorial