Setting Up Assisted Annotation

Assisted annotation enables rapid data labeling, improving efficiency and accuracy while freeing your team to focus on critical tasks.

Using Large Language Models Based Assistance

Rely on zero-shot and few-shot learning approaches to annotate your data:

• Zero-shot learning: Models identify entities without prior labeled examples.

• Few-shot learning: Models learn from a small dataset with 1–5 labeled examples.

Steps to Add an LLM Model

• Navigate to the drop down menu next to the predict button in the annotation interface of a document then select configure prediction.

• Select Task and select LLM

• select a pretrained model and version (default: OpenAI's GPT-3).

• Configure settings:

  • Number of Few shots examples (0–5).

  • Context length

  • Temperature

  • Add the lables you need help with and their description.

• Press Save to finalize.

Hugging Face Models Based Assistance

UbiAI integrates with Hugging Face, allowing you to use thousands of pre-trained models for auto-labeling.

Steps to Add a Hugging Face Model

• Go to the Profile page in UbiAI and enter your Hugging Face API key.

• Navigate to the drop down menu next to the predict button in the annotation interface of a document then select configure prediction. Select Task and select Hugging Face.

  Click Add New Model and provide the following details:

  • Model Name: A descriptive name for your model.

  • API_URL Link: The endpoint link for the Hugging Face model you want to use.

• Press Create Model to finalize the setup then use the model to annotate your data automatically.

Pretrained Models Based Assistance

You can use your own pre trained models for predictions on UbiAI

Steps to Add a pretrained Model

• Navigate to the drop down menu next to the predict button in the annotation interface of a document then select configure prediction. Select Task and select model.

  Click Add New Model and provide the following details:

  • Model Name: the model name as saved in UbiAI.

• Press Create Model to finalize the setup then use the model to annotate your data automatically (click "predict").