# Need Suggestion

**URL:** <https://discuss.huggingface.co/t/need-suggestion/81570>\
**Category:** Research\
**Created:** [April 13, 2024, 3:06pm UTC](https://discuss.huggingface.co/t/need-suggestion/81570 "2024-04-13T15:06:35Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![khurramnaseem](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/khurramnaseem/32/25800_2.png) [@khurramnaseem](https://discuss.huggingface.co/u/khurramnaseem)\
**Post date:** [April 13, 2024, 3:06pm UTC](https://discuss.huggingface.co/t/need-suggestion/81570/1 "2024-04-13T15:06:35Z")

</div>

Hi all,

I’m new to this LLM’s world and I need suggestions on the following idea. I want to fine-tune a LLM model based on exam past papers, which are currently available on PDF format.

**Objective** : The model understand and explain past answers and able to generate new questions and answers.

1. Which LLM model I should select for this purpose? Keep in mind I want to start from very basic as I mentioned earlier I’m very novice.

2. The past papers are in PDF format with pictures as well, Do I need to convert them on some specific format like JSON?

---

<div class="post-metadata">

**Author:** ![PervaizKhan](https://avatars.discourse-cdn.com/v4/letter/p/85f322/32.png) [@PervaizKhan](https://discuss.huggingface.co/u/PervaizKhan)\
**Post date:** [April 18, 2024, 10:19pm UTC](https://discuss.huggingface.co/t/need-suggestion/81570/2 "2024-04-18T22:19:54Z")

</div>

Hi,

You may look at huggingface llms leaderboard (search on google for exact link), and try to pick the latest models for your tasks.  
For pdf, I am not sure, but I guess you may convert them to text format some , maybe in csv format or json etc.  
For finetuning llm, please look at the topics such parameter efficient finetunig (PEFT).

I am nit exactly sure but you may also look at “vision llm” that works on images and text.  
I hope this helps.  
Good luck.

---

<div class="post-metadata">

**Author:** ![khurramnaseem](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/khurramnaseem/32/25800_2.png) [@khurramnaseem](https://discuss.huggingface.co/u/khurramnaseem)\
**Post date:** [April 19, 2024, 11:45am UTC](https://discuss.huggingface.co/t/need-suggestion/81570/3 "2024-04-19T11:45:02Z")

</div>

Thank you @PervaizKhan for your reply.