Blogs are boring. Things have changed we are close to AGI but, still the blogs are from ancient days with no quality of life improvement. So, I have made this repository as a framework or guideliness to create a new way of interactive blog.

Specially for mathematics blog, it would be great if you could tweak the variable of a formula and see the result. It will help you see and test the edge cases. This was my main motivation.

This project uses modular architecture to help you create interactive blog instantly and here are its cool features;

• Supports markdown formatting, dash doesn't let you do so but I made this possible.
• Utility classes for styling that covers most of the portion so, you can make the site look cool.
• Utility classes that are similar to that of TailWind CSS to help you catchup easier.
• You can create interactive sections in between with visualization through plotly.

1. dash provides the Python to React feature,
2. plotly helps me to build interactive Graphs easily.

I have created an example blog of visualizing LLM to let you know the truepotential of the interactive blog.

1. Simple architecture for understanding,
2. 124M parameter model that can run easily without the need of dedicated GPU,
3. First of its kind so, a good start to know where thing started.

To visualize the embeddings space of tokens, I used GPT-2 embedding matrix to create a lookup dictionary such that whenever user enters a token they can compare the similarlity in the vector space.

Here, to visualize the things better we have also converted the 768 dimensional vector space into 3-d space using PCA. This is to create a 3d plot using plotly and show that to the user.