Skip to content

Cappy

A small scorer that boosts large multitask language models

Visit official site
Category AI models

What Cappy is

Cappy is described as a lightweight pretrained scorer built on RoBERTa that improves the performance and efficiency of large multitask language models. It can stand alone or support a language model.

What you can do with it

  • Solve classification tasks independently
  • Act as an auxiliary component for language models
  • Fine-tune on downstream tasks to add supervision

Application scenarios

  • A customer service bot uses it for dialogue quality
  • A search engine uses it for result relevance
  • Summarisation and text generation outputs are optimised

Pros & cons

✓ What we like

  • Works with both open source and closed source models
  • Can be used alone for classification tasks

! What to watch out for

  • The scorer is tied to RoBERTa as its base
  • Performance gains depend on downstream fine-tuning

FAQ

How large is it?

The material states 360 million parameters.

Does it need backpropagation?

The material says it does not need backpropagation into the language model parameters.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review