SliceX AI Models
Overview
The SliceX AI™ Cloud Platform currently gives access to different proprietary model families optimized for different NLP tasks and deployment scenarios. All models can be trained from scratch or initialized with pre-trained weights and fine-tuned.
| Model Family | Description & Target Use-cases |
|---|---|
| Papaya | Models that can analyze natural language text and make predictions. Use Cases: text classification, content moderation, topic categorization … |
| Grapefruit | Models that can analyze natural language text or conversational data and make predictions (or selections) about part of the text. Use Cases: sequence labeling, span prediction, text extraction … |
| Dragonfruit | Models that can find relevant content given an input query in natural language. Use Cases: question answering, semantic search, content recommendation… |
| Jackfruit | Models that can encode natural language text and generate a vector representation that captures its meaning. Use Cases: text embedding, query embedding, user embedding … |
We are continuously improving our existing suite of models as well as developing new ones. Keep a lookout for new models (e.g., generation) in the works to be released soon!
Choosing the Model
When training a custom model, the SliceX AI™ Cloud Platform automatically selects the best model family based on the task type and sets it as the default option in the model configuration.
Model Size
All families of SliceX AI proprietary models can be configured in different sizes suited for different deployment scenarios.
There are 3 model sizes available: Mini, Base, Large (upcoming)
This choice of model size enables flexible optimization choices for deploying SliceX AI models depending on customer application, device and hardware. For example, a text classification app can customize the regular model (Papaya) for Cloud deployment and choose to deploy the smaller version (PapayaMini) on the Edge.
Example
Say you want to train a model for named entity recognition. The best choice of model family for this task is Grapefruit. In order to you start a training job, you need use the SliceX AI Trainer and fill in the appropriate model configuration:
{
"model_config": {
"name": "ner-model",
"type": "sequence-labeling",
"family": "Grapefruit",
"size": "mini",
},
}