Pre-trained models
Deploy a pre-trained model to kickstart your ML-powered application.
Baseten provides a growing set of pre-trained models that solve common ML tasks. These models are a great way to kickstart your ML application and showcase the features and functionality of Baseten - you can deploy pre-trained models along with optional application templates to add ML into your application in minutes.
Workspaces on the free Personal plan are limited to a single deployed model. Pre-trained models provided by Baseten do not count against this limit. To increase the number of models for your workspace, upgrade to a paid Baseten plan.
Pre-trained models offer many of the same features as your own models deployed on Baseten:
  • You can invoke the model directly through the Baseten Python client or an API call
  • You can create an application for the model
  • You can use the model in worklets
  • They come with editable readmes
Pre-trained models have some limitations that your models do not have:
  • You can't view health metrics or logs for pre-trained models
  • Pre-trained models do not have multiple versions
  • Pre-trained models have fixed model resources and auto-scaling settings
  • Pre-trained models do not count against the models limit in your workspace


To deploy a pre-trained model, head to the Explore tab on the Models page. Scroll through pre-trained models, filter by tags, or search for the model you want to deploy. Click on that model,
Deploying a pre-trained model
By default, every pre-trained model comes with a starter application demonstrating how to use the model for a common use case. When you're in the deployment modal, you can opt out of creating this starter application by toggling off the "Create a starter application" option.
Once the model is deployed, you can edit the starter app or use the model in your applications.

Available models

Pre-trained models can be applied to common ML tasks. From speech transcription to sentiment analysis, image classification to photo restoration, pre-trained models deliver powerful ML capabilities directly to your applications. Baseten currently offers 23 pre-trained models:
  • CLIP: connecting text and images (Custom): Classify images with zero-shot-like functionality.
  • CodeGen mono 2B (Custom): Generate Python code from natural language or code prompts. This version of the model was fine-trained on Python code.
  • CodeGen multi 2B (Custom): Generate Python code from natural language or code prompts. This version of the model was fine-trained on natural language and broad programming languages.
  • Dall·E Mini (Custom): Generate novel images from a text prompt.
  • ELMo word representations (TensorFlow): Generate embeddings from a language model trained on the 1 Billion Word Benchmark.
  • English to French translation (Hugging Face: Transformer): Translate between English and French.
  • Extractive question answering (Hugging Face: Transformer): Extract an answer from a given text when provided a question.
  • Faster R-CNN Inception V2 (TensorFlow): Object detection model using Faster R-CNN with Inception
  • GFP-GAN (Custom): Restore photos with GFP-GAN.
  • GPT-J (Custom): Generate text with GPT-J. This is an implementation of EleutherAI GPT-J-6B.
  • Image Segmentation (Custom): Identify classes of objects in an image.
  • Iris random forest classifier (scikit-learn): Predict Iris class with a random forest classifier.
  • Masked language modeling (Hugging Face: Transformer): Fill a masked token in sequences based on the context around it.
  • MNIST digit classifier (scikit-learn): Logistic Regression model for classification of MNIST database of handwritten digits.
  • PDF Invoice parser (Google): A PDF parsing model that extracts structured data from invoices.
  • ResNet50 V2 (TensorFlow): Image detection model using ResNet V2 neural network architecture.
  • Sentiment analyzer (Hugging Face: Transformer): Analyze sentiment.
  • Style transfer (Custom): Apply one image's style onto another.
  • Summarizer (Hugging Face: Transformer): Summarize a text into a shorter text.
  • Text generation (Hugging Face: Transformer): Generate language given a prompt.
  • Token classification (Hugging Face: Transformer): Classify tokens in strings.
  • Wav2vec 2.0 speech transcription (Custom): Transcribe audio files with wav2vec 2.0.
  • Zero-shot classification (Hugging Face: Transformer): Classify snippets of text into unseen categories.
If you'd like Baseten to offer a specific pre-trained model, or if you'd like to fine-tune these models on your data, reach out.
Copy link
Available models