Refuel Updates (Dec '24)

👋 We’re back with another edition of Refuel updates!

Here’s what’s new with Refuel from this past month:

1. Refuel-LLM V2 Mini

Refuel-LLM V2-Mini is Refuel’s newest and smallest 1.5B parameter model. Trained using the same recipe as other models in Refuel-LLM model family, Refuel-LLM-2-Mini (74.84) outperforms other open source models in its weight range, such as Alibaba's Qwen-1.5B-Instruct (54.71), Google's Gemma-2-2B-it (67.83) and Meta's Llama-3.2-3B-Instruct (56.00) on our previously shared benchmark of data labeling, extraction, enrichment and matching tasks. Users can use this base model to fine tune models with higher throughput and lower cost compared to larger LLMs.

2. Improved support for images and documents in application requests

Refuel now offers better native support for processing images and documents, including:

1. "Text Recognition" enrichment to extract text from images and documents.

2. Ability to pass images and documents (or equivalently, URIs to these in your cloud storage buckets) as inputs directly for VLMs.

3. Javascript SDK (beta)

The JavaScript SDK is a lightweight client library for interacting with the Refuel’s APIs. Leverage the SDK to transform data, manage application responses, and collect feedback seamlessly within your JavaScript/TypeScript applications.

Bonus:

• We wrote about analyzing call transcripts with Refuel

• pip install --upgrade refuel to get the latest features and usability improvements.

• Latest docs available here

Let us know if you have questions or feature suggestions – enjoy the holidays and see you next month!