# Finetuning on base or instruct model?

**URL:** <https://discuss.huggingface.co/t/finetuning-on-base-or-instruct-model/80416>\
**Category:** Beginners\
**Created:** [April 6, 2024, 5:23am UTC](https://discuss.huggingface.co/t/finetuning-on-base-or-instruct-model/80416 "2024-04-06T05:23:57Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![mooolie](https://avatars.discourse-cdn.com/v4/letter/m/6de8d8/32.png) [@mooolie](https://discuss.huggingface.co/u/mooolie)\
**Post date:** [April 6, 2024, 5:23am UTC](https://discuss.huggingface.co/t/finetuning-on-base-or-instruct-model/80416/1 "2024-04-06T05:23:57Z")

</div>

Hi, i want to finetune [this model](https://huggingface.co/upstage/SOLAR-10.7B-Instruct-v1.0), should i use the base model or the instruct fine tuned model?

I’ve seen some voices saying that we should fine tune on the base model so that the data used to fine tune the instruct model would not contaminate my dataset or mess with the prompt format that i’m using?

Can anyone clarify this issue for me? Thank you.

---

<div class="post-metadata">

**Author:** ![taki555](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/taki555/32/54675_2.png) [@taki555](https://discuss.huggingface.co/u/taki555)\
**Post date:** [October 9, 2025, 2:36pm UTC](https://discuss.huggingface.co/t/finetuning-on-base-or-instruct-model/80416/2 "2025-10-09T14:36:06Z")

</div>

Try to tune on Base, but execute on Instruct. Check this: [Paper page - Shadow-FT: Tuning Instruct via Base](https://huggingface.co/papers/2505.12716)