How to beginner · 3 min read

How to load a pretrained model from Hugging Face

Q: How to load a pretrained model from Hugging Face

Use the transformers library to load a pretrained model from Hugging Face by calling AutoModel.from_pretrained() and AutoTokenizer.from_pretrained() with the model name. This loads the model weights and tokenizer for inference or fine-tuning.

Quick answer

Use the transformers library to load a pretrained model from Hugging Face by calling AutoModel.from_pretrained() and AutoTokenizer.from_pretrained() with the model name. This loads the model weights and tokenizer for inference or fine-tuning.

PREREQUISITES

Python 3.8+
pip install transformers
pip install torch (or tensorflow)
Internet connection to download model weights

Setup

Install the transformers library and a backend deep learning framework like torch or tensorflow. Set up your Python environment with these commands:

bash

pip install transformers torch

Step by step

Load a pretrained model and tokenizer from Hugging Face by specifying the model name. This example uses bert-base-uncased for demonstration.

python

from transformers import AutoModel, AutoTokenizer

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

# Tokenize input text
inputs = tokenizer("Hello, Hugging Face!", return_tensors="pt")

# Forward pass
outputs = model(**inputs)

print(outputs.last_hidden_state.shape)  # torch.Size([1, 5, 768])

output

torch.Size([1, 5, 768])

Common variations

Use AutoModelForSequenceClassification to load models fine-tuned for classification tasks.
Switch backend to TensorFlow by installing tensorflow and passing from_pt=True if needed.
Load models from local directories by passing the path instead of a model name.

python

from transformers import AutoModelForSequenceClassification

# Load a classification model
model = AutoModelForSequenceClassification.from_pretrained("distilbert-base-uncased-finetuned-sst-2-english")

Troubleshooting

If you see a Model name not found error, verify the model name spelling or check availability on Hugging Face Hub.
For slow downloads, consider caching models locally by specifying cache_dir in from_pretrained().
If you get CUDA errors, ensure your PyTorch installation matches your GPU setup.

✅

Key Takeaways

Use the transformers library's Auto classes to load pretrained models and tokenizers easily.
Specify the exact model name from Hugging Face Hub to download weights automatically.
You can load models for different tasks by choosing the appropriate AutoModel variant.
Local caching and backend framework choice help optimize loading and inference.
Check model availability and environment compatibility to avoid common errors.

Verified 2026-04 · bert-base-uncased, distilbert-base-uncased-finetuned-sst-2-english

Verify ↗