zoom_in Click to enlarge

Orca 2

▲ ▼

description Orca 2 Overview

Orca 2 is an open-source small language model developed by Microsoft Research and released in late 2023. It is built by fine-tuning Meta's Llama 2 base models, specifically the 7-billion and 13-billion parameter versions. The training methodology focuses on teaching the model cautious reasoning, enabling it to choose between different processing strategies—such as step-by-step reasoning or direct answers—based on the specific task. It is intended for developers who need capable, lightweight models for local deployment and complex reasoning tasks.

help Orca 2 FAQ

What is Orca 2?

Orca 2 is a small language-model family developed by Microsoft Research and released in late 2023. It was designed to improve reasoning behavior in models that are much smaller than frontier systems.

Which model sizes does Orca 2 include?

The Orca 2 release includes 7-billion-parameter and 13-billion-parameter versions. These are commonly referred to as Orca-2-7B and Orca-2-13B.

What is Orca 2 based on?

Orca 2 was fine-tuned from Meta's Llama 2 base models, including the 7B and 13B sizes. Its training focused on teaching smaller models different reasoning strategies rather than simply copying final answers.

Why was Orca 2 created?

Microsoft Research created Orca 2 to study whether smaller language models could learn stronger reasoning behavior from carefully generated training data. The research compared the 7B and 13B models with larger Llama 2 systems on reasoning benchmarks.

swap_horiz

Looking for Orca 2 alternatives? Compare top competitors ranked & scored

arrow_forward