At Baseten, we’re committed to maximizing the chances of success of every machine learning project. A big aspect of this effort is making it as easy as possible to go from zero-to-one on a project. That’s why we built the Baseten model zoo. Get your project off to a flying start by using a state-of-the-art pre-trained model as your initial benchmark.
The Baseten model zoo
Today, the Baseten model zoo includes 17 state-of-the-art models that solve common image, text, and audio tasks and can be used in projects right away. You can deploy these models and start using them in applications in a matter of minutes. Really! Check out this content moderation app built using the pre-trained zero-shot classification model from the Baseten model zoo.
Here are a few of the pre-trained models that we’re particularly excited about:
- Wav2vec for audio transcription: A version of wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representation implemented on the huggingface transformer library. It’s been learned using self-supervision and fine-tuned on transcribed speech.
- Zero-shot classification for text classification: This model solves for the task of classifying text into unseen categories using the hugging-face zero-shot-classification pipeline.
- GPT-J for language tasks: The famed OpenAI text model with a general-purpose “text-in, text-out” interface, which makes it possible to apply it to virtually any language task.
Use a model zoo model in an application as a baseline to improve upon. Or, you might even find that one of these models is sufficient to solve the problem you had in mind— for example, a large consumer tech startup is using the Baseten model zoo for a complex content moderation workflow involving detecting hate speech in audio files uploaded to their platform.
Ready, set, go!
Model zoo models can be deployed and served using Baseten with just a few clicks and come with all the bells and whistles of our model hosting product (including support for inference with GPUs).
Once deployed, these models can be embedded within worklets to be used as APIs or within user-interfaces built in Baseten. To make getting started even easier, when you deploy a pre-trained model, you can choose to have Baseten automatically create a sample application and use that to test the model or build off of for your own application.
See our documentation for more information.
If you have a Baseten account, then the full model zoo (including pre-trained Hugging Face, TensorflowHub, and OpenAI models) is available to you. We’re also adding new models every month, so if there’s a specific model you’d like to see in Baseten, let us know.
Fine-tune FLAN-T5 on Blueprint today!
You can now fine-tune FLAN-T5, an instruction-tuned text-to-text transformer model developed by Google, on Blueprint!