---
title: "Discussion on \"Deep dive into KoboldAI: The AI you can run locally\" | Hashnode"
description: "Discussion on \"Deep dive into KoboldAI: The AI you can run locally\". Discover the world of text-to-text generation with KoboldAI with this  article. We'll go through how to setup KoboldAI and use it in its various modes."
url: "https://hashnode.com/posts/deep-dive-into-koboldai-the-ai-you-can-run-locally/63f27d5fd52813c9dd7ad6e1"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [Deep dive into KoboldAI: The AI you can run locally](https://blog.tarekjellali.com/deep-dive-into-koboldai-the-ai-you-can-run-locally?utm_source=hashnode&utm_medium=feed)

Discover the world of text-to-text generation with KoboldAI with this article. We'll go through how to setup KoboldAI and use it in its various modes.

Feb 19, 2023 · 12 min read · blog.tarekjellali.com  
Read the article

-   [#ai](https://hashnode.com/tag/ai)
-   [#chatgpt](https://hashnode.com/tag/chatgpt)
-   [#artificial-intelligence](https://hashnode.com/tag/artificial-intelligence)
-   [#koboldai](https://hashnode.com/tag/koboldai)

## Written by

[Tarek Jellali](https://hashnode.com/@tarek-jellali)

@tarek-jellali · Paris, France

  
[View profile](https://hashnode.com/@tarek-jellali)

Software Engineer

Software Engineer • TypeScript • JavaScript • Angular • VueJS • WordPress • Ex-founder of ng-enious

## Responses 2

Markdown supported · ⌘↵ to post

[Tarek Jellali](https://hashnode.com/@tarek-jellali)

[Mar 25, 2023](https://hashnode.com/posts/deep-dive-into-koboldai-the-ai-you-can-run-locally/63f27d5fd52813c9dd7ad6e1/comment/641e4081ecc310467c900b48)

[Henky!!](https://hashnode.com/@henk717) thank you so much for the comment, honestly your project is worth more attention especially because of its amazing quality!

I'll make sure to add your precisions to the article, thank you again!

[Henky!!](https://hashnode.com/@henk717)

[Mar 24, 2023](https://hashnode.com/posts/deep-dive-into-koboldai-the-ai-you-can-run-locally/63f27d5fd52813c9dd7ad6e1/comment/641dd4a105c482464c909e86)

Fun to see our project being featured on your blog! I do want to add that the 125M models are usually only considered useful for quick feature testing. They are very incoherent and will not properly showcase all the features we have.

Adventure mode for example is specifically designed for the models in our Adventure Model category since the base models do not have suitable training for the mode.

Home users often can run 2.7B locally which while lacking on logical reasoning provides good enough results for stuff like a basic text adventure or story if the user provides assistance by correcting mistakes.

We offer an online Google Colab notebook where people can try models for free using one of Google's computers, which is recommended for people who can not run them at home. The links for this are inside the readme (Currently the TPU edition has been broken by Google).

I hope this helps people willing to try it out to get a better experience from the models.
