Explain more about "finetunable" attribute on Jumpstart model card

This doc https://docs.aws.amazon.com/sagemaker/latest/dg/jumpstart-fine-tune.html says only models with 'finetunable" can be finetuned. But I found that for mistral-7b, which is tagged as not "finetunable", can be finetuned in several tutorials (e.g. https://www.datacamp.com/tutorial/mistral-7b-tutorial).

Can you explain more about this attribute on the model card?

주제

기계 학습 및 AI

태그

아마존 SageMaker 아마존 세이지메이커 점프스타트

언어

English

Spring

질문됨 4달 전1838회 조회

3개 답변

최신
최다 투표
가장 많은 댓글

수락된 답변

Hi,

not finetunable in Sagemaker JumpStart doesn't mean that a model cannot be finetuned in general. It only means that this same model cannot be finetuned in the way that it has been packaged in JumpStart.

So, you can fine tune Mistral in Sagemaker if you load it from another way than Sagemaker JumpStart.

Best,

Didier

전문가

Didier_Durand

답변함 4달 전

Thanks a lot, Didier.

Spring

답변함 4달 전

Didier_Durand 전문가
4달 전
Happy to have helped! If the answer is satisfactory, please accept it to keep the ball rolling
Didier_Durand 전문가
4달 전
Thanks to you for accepting the answer

Can I additionally know which version of mistralai/Mistral-7B-Instruct is being used in Jumpstart? Mistral now has mistralai/Mistral-7B-Instruct-v0.2.

Spring

답변함 4달 전

Didier_Durand 전문가
4달 전
Look at https://aws.amazon.com/blogs/machine-learning/mistral-7b-foundation-models-from-mistral-ai-are-now-available-in-amazon-sagemaker-jumpstart/

It will tell you how to discover all Mistral models available in JumpStart

Explain more about "finetunable" attribute on Jumpstart model card

관련 콘텐츠