Language model assistance

I’m surprised nobody did it yet, but it could be cool to have autocompletion in Obsidian.

As a programmer, GitHub Copilot increase a lot my productivity and it would be the same for writing stuff.

A plugin could do either edge autocompletion using a small model, Android have such simple models on TensorflowLite. On desktop could have something more powerful maybe.

But very simple thing could be to use GPT3, but I know not everybody has access to it.
The Codex models are free to use and works well even for normal language even though it’s supposed to be used for code.

There is also HuggingFace API which offer free characters per month.
API usage could enable to use large language models.

6 Likes

It’s a good idea, but again it’s a question of means and processing power.

To do something like this, we need to train an algorithm on all the content of our vault, which presupposes that there is already enough content.

However, this requires powerful machines, which is not available to everyone, and this is the problem that makes it impossible to implement (for the moment) as a plugin.

Even using free services, it’s not “sustainable” on the long term (especially when I see how fast my vault grows on a daily basis).

For the mere inference, running a CLIP model with ONNX is interesting

Otherwise, with an API model:

Early version that works, but UX is not great

Added a notebook to fine-tune GPT3 on your vault obsidian-ava/openai_fine_tune.ipynb at main · louis030195/obsidian-ava · GitHub

2 Likes

You can try Text Generator Plugin which uses GPT-3.

1 Like

GitHub - louis030195/obsidian-ava: Thinking assistance using language model/multimodal in https://obsidian.md now supports stable diffusion

1 Like

I updated GitHub - louis030195/obsidian-ava: AI-enhanced reflection in Obsidian https://obsidian.md to show how to use with GitHub - SilentVoid13/Templater: A template plugin for obsidian. You can create your own prompts and prompt engineering logic to write in Obsidian, generate code, generate prompt + stable diffusion images, etc.

What is the difference between the text generator plugin and the ava genrator? ELI5 I have only very basic knowlegde of GPT-3 and things.

The text generator plugin only generates text, while the ava generator mainly generates images. Both work like autocomplete on superpowers. Here’s a good demo of the text generator plugin: Writing a small blog Article using AI in less than 5 minutes - YouTube

1 Like

Why? It generates text equally, maybe you didn’t get it?

For me it’s a question of convenience. Right now I’ve switched to vscode to use the copilot extension. If same plugin existed for obsidian that would be very helpful.

2 Likes