Comment on page

Project Models

UBIAI empowers users to create and train custom models for various natural language processing tasks, enabling efficient labeling. The models you can fine-tune includes NER, Relation Extraction, Span categorization and text classification.

Named Entity Recognition

Custom Models

You can select a custom trained Named Entity Recognition (NER) models tailored to your specific domain or task. The tool allows for easy annotation of labeled entities in text data. The process involves:
To use the model for auto-labeling, select it(only one model at a time can be selected) from the list of available models. The actual model can be trained from the Model Menu in the navigation bar, refer to the Model-Assisted Labeling section.

LLM

Leverage Large Language Models (LLMs) for NER with zero-shot and few-shot learning approaches. Zero-shot learning enables the model to recognize entities without explicit training, while few-shot learning involves training the model on a small dataset with a few labeled examples.
To add an LLM model, simply click on "Add new model", by default OpenAI's GPT 3.5 Turbo will be added. You can configure the number of examples you would like to provide to GPT from 0 to 5. For more information, please refer to the Zero-shot and Few-shot Labeling section.

Huggingface Models

Select from the thousands of pre-trained models in the Hugging Face (HF) Transformers library to auto-label your data.
Simply press on "Add new model", input a name for your model and its HF URL as show below.

Relation Extraction

Custom Models

Train relation extraction models to extract relationships between two entities in text data.
To use the model for auto-labeling, select it (only one model at a time can be selected) from the list of available models. The actual model can be trained from the Model Menu in the navigation bar, refer to the Model-Assisted Labeling section.

LLM

Coming soon...

Span Categorizer

Custom Models

Span Categorization involves labeling specific spans of text such as long sentences or paragraphs with predefined label. UBIAI facilitates training span categorizer models to identify and categorize spans.
To use the model for auto-labeling, select it (only one model at a time can be selected) from the list of available models. The actual model can be trained from the Model Menu in the navigation bar, refer to the Model-Assisted Labeling section.

Text Classification

Custom Models

Train models to classify entire documents into predefined categories. To use the model for auto-labeling, select it (only one model at a time can be selected) from the list of available models. The actual model can be trained from the Model Menu in the navigation bar, refer to the Model-Assisted Labeling section.

LLM

Leverage Large Language Models (LLMs) for text classfication with zero-shot and few-shot learning approaches. Zero-shot learning enables the model to recognize categories without explicit training, while few-shot learning involves training the model on a small dataset with a few labeled examples.
To add an LLM model, simply click on "Add new model", by default OpenAI's GPT 3.5 Turbo will be added. You can configure the number of examples you would like to provide to GPT from 0 to 5. For more information, please refer to the Zero-shot and Few-shot Labeling section.