# Finetune generative model

**URL:** <https://discuss.huggingface.co/t/finetune-generative-model/32927>\
**Category:** Beginners\
**Created:** [March 2, 2023, 11:11pm UTC](https://discuss.huggingface.co/t/finetune-generative-model/32927 "2023-03-02T23:11:52Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![welu](https://avatars.discourse-cdn.com/v4/letter/w/f1d935/32.png) [@welu](https://discuss.huggingface.co/u/welu)\
**Post date:** [March 2, 2023, 11:11pm UTC](https://discuss.huggingface.co/t/finetune-generative-model/32927/1 "2023-03-02T23:11:52Z")

</div>

I want to train a model on my custom documentation in a way it is able to answer related questions like chatgpt style (generative way). But I am lost. What is the best method to do this and what is the best model for the task.

---

<div class="post-metadata">

**Author:** ![gaganluthra](https://avatars.discourse-cdn.com/v4/letter/g/f14d63/32.png) [@gaganluthra](https://discuss.huggingface.co/u/gaganluthra)\
**Post date:** [March 7, 2023, 4:36pm UTC](https://discuss.huggingface.co/t/finetune-generative-model/32927/2 "2023-03-07T16:36:47Z")

</div>

I am a total beginner too. However, I will try my best.

I tried to solve a similar problem as you, and have tried a few things.

So far I am unable to make any opensource model work for this (or in general).

However, I saw somewhere in OpenAI API docs that you can upload documents and their models will answer questions from your documentation.

I have successly been able to use davinci model playground to answer questions based on the context (included in the prompt). In their playground there is a “chat” preset. I used that and modified it to my use case.

I havent deployed it anywhere yet, but as an experiment my mind was almost blown!

Because we both are working on a similar problem, I would like to stay updated on your progress too.

---

<div class="post-metadata">

**Author:** ![welu](https://avatars.discourse-cdn.com/v4/letter/w/f1d935/32.png) [@welu](https://discuss.huggingface.co/u/welu)\
**Post date:** [March 9, 2023, 10:34pm UTC](https://discuss.huggingface.co/t/finetune-generative-model/32927/3 "2023-03-09T22:34:53Z")

</div>

Sure. I like the idea of keep updating each other.  
For the moment I think this is the best approach I have found : [google/t5-11b-ssm-tqa · Hugging Face](https://huggingface.co/google/t5-11b-ssm-tqa?text=When+was+Franklin+D.+Roosevelt+born%3F)  
Finetune this model would give us the QA model we need. But I need to dive deeper into the documentation.