https://github.com/maxanier/llm_editor

https://github.com/maxanier/llm_editor

Last synced: about 1 month ago
JSON representation

• Host: GitHub
• URL: https://github.com/maxanier/llm_editor
• Owner: maxanier
• License: mit
• Created: 2024-08-13T09:50:28.000Z (3 months ago)
• Default Branch: main
• Last Pushed: 2024-08-19T20:51:58.000Z (3 months ago)
• Last Synced: 2024-08-20T00:28:13.719Z (3 months ago)
• Language: Python
• Size: 10.7 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# LLM Text Editor

This script sends text chunks to an `ollama` server running the Large Language Model (LLM).

It performs pre- and post-processing and creates a diff between the original text and the edited version.

Not perfect, but helpful.

## Usage

1. You need an [ollama](https://github.com/ollama/ollama/) instance running. Adjust the end-point if necessary

2. Make sure you have to desired model installed

3. Save the text you want to check in `inout/input.txt`. (See `inout/input.example.txt`)

4. If it is a long text, add lines starting with `#` to break the text into chunks. (E.g. preprend headings with `#`)

5. Run script (see `python main.py --help`)

6. e.g. run [diff-so-fancy](https://github.com/so-fancy/diff-so-fancy) on the output diff file (`cat inout/diff.txt | diff-so-fancy`)

## Models

Different models and corresponding system prompts can be configured in `llm_config.json`.

Most concise results are achieved with `karen` so far.

## Latex

If you are writing a latex article and want to process the text, you need to somehow extract it from there.

I haven't found a good way yet. 

My current approach is the following:

1. Create a PDF that is as "clean" as possible (no figures, no equations, single-column). See `/latex` for this.

2. Use `extract_pdf.py` to get the text

3. Some manual fine-tuning