Orca-2

Affiliation

Microsoft

Commercial

Fine-tuning Method

SFT

Note

tailored, high-quality synthetic data generated by GPT4 • intuition : small capa에서는 large capa에서의 생성 전략이 best가 아닐 수 있음. GPT-4가 direct answer를 내지만, smaller model은 더 작은 task 단위로 나눠서 답변을 생성해야함. • how-to : very detailed instructions and even multiple calls

데이터

모델 크기

7B 13B

새롭게 제공된 Resource

Model

출시일

2023-11

•

최근 MS 내에서 SLM (small-scaled LM) 연구가 활발하게 진행되고 있는데, 대부분의 approach가 data organization & synthetic data generation을 통한 성능 향상

Orca-2

7B, 13B
23/11
fine-tuning LLAMA2 on tailored, high-quality synthetic data.

•

ai-platform intro : https://techcommunity.microsoft.com/t5/ai-ai-platform-blog/continuing-to-advance-state-of-the-art-model-and-tooling-support/ba-p/4008407

•

blog : https://www.microsoft.com/en-us/research/blog/orca-2-teaching-small-language-models-how-to-reason/

•

orca-2 : https://www.microsoft.com/en-us/research/publication/orca-2-teaching-small-language-models-how-to-reason/

◦

gpt-4 synthetic data (an expanded, highly tailored synthetic dataset) : various reasoning techniques (step-by-step processing, recall then generate, recall-reason-generate, extract-generate, and direct answer methods) & teach it to choose the most effective solution strategy for each task

◦

intuition : small capa에서는 large capa에서의 생성 전략이 best가 아닐 수 있음. GPT-4가 direct answer를 내지만, smaller model은 더 작은 task 단위로 나눠서 답변을 생성해야함.

◦

how-to : very detailed instructions and even multiple calls

◦

result : 13B 사이즈로 llama-2-chat-70B를 능가하는 NL understanding 성능

Reference

•

orca-1 : https://www.microsoft.com/en-us/research/publication/orca-progressive-learning-from-complex-explanation-traces-of-gpt-4/

◦

gpt-4 synthetic data (large-scale and diverse imitation data with judicious sampling and selection) :  GPT 4 including explanation traces; step-by-step thought processes; and other complex instructions, guided by teacher assistance from ChatGPT

License

https://huggingface.co/microsoft/Orca-2-13b/blob/main/LICENSE

huggingface.co

https://huggingface.co/microsoft/phi-2/resolve/main/LICENSE

These license terms are an agreement between you and Microsoft Corporation (or one of its affiliates). They apply to the source code, object code, machine learning models, or data (collectively “Materials”) that accompany this license. IF YOU COMPLY WITH THESE LICENSE TERMS, YOU HAVE THE RIGHTS BELOW. BY USING THE MATERIALS, YOU ACCEPT THESE TERMS. 1) INSTALLATION AND USE RIGHTS TO THE MATERIALS. Subject to the terms of this agreement, you have the below rights, if applicable, to use the Materials solely for non-commercial, non-revenue generating, research purposes: