Skip to main content

Overview

The SliceX AI™ Cloud Platform gives you the ability to train the SliceX AI proprietary models on your custom datasets. We also make available several configurable options for specifying model parameters. Refer to the training job customization section for more details.

Docs Version Dropdown

Data Privacy/Security​

Every dataset that is uploaded in the SliceX AI Trainer is automatically deleted after a training job is completed or canceled. The SliceX AI™ models are also privacy preserving and encrypted with access keys.

SliceX AI Models​

The SliceX AI™ Cloud Platform currently gives access to 3 proprietary model families: Papaya, Dragonfruit and Grapefruit. All models can be trained from scratch or initialized with pretrained weights and fine-tuned on downstream datasets. They are available in 2 sizes, mini or base. You need to specify the model family and size in your training requests. As of now:

  • Papaya can be used for text classification
  • Dragonfruit can be used for question answering
  • Grapefruit can be used for sequence labeling

Supported Tasks​

Text classification (binary, multiclass and multilabel), sequence labeling and question answering are currently supported for custom training. Question answering also comes equipped with the option to generate embeddings from an existing model, using a new set of passages.

Custom training for each task type has specific but well defined dataset format requirements. Review the specifications here under Dataset Requirements.

Custom Inference​

Once a model is trained, it is immediately available for inference (check out the Custom Inference section).

Authentication​

Similar to the inference APIs, the SliceX AI Trainer uses API keys for authentication. Please visit the Quickstart tutorial for a review of the sign up and API access process. You will need the API keys for all your requests- they must be added as a parameter in the header for every training request. Do not share your keys with others or expose them in client-side code! We recommend that you keep your API keys securely loaded in your backend server.