https://github.com/attentionmech/dex

pokedex for LLMs
https://github.com/attentionmech/dex

architectures interactive interactive-visualizations large-language-models llms machine-learning models visualization webapp

Last synced: 2 months ago
JSON representation

pokedex for LLMs

• Host: GitHub
• URL: https://github.com/attentionmech/dex
• Owner: attentionmech
• License: apache-2.0
• Created: 2025-04-04T02:03:25.000Z (12 months ago)
• Default Branch: main
• Last Pushed: 2025-04-14T10:13:56.000Z (12 months ago)
• Last Synced: 2025-04-14T11:23:59.707Z (12 months ago)
• Topics: architectures, interactive, interactive-visualizations, large-language-models, llms, machine-learning, models, visualization, webapp
• Language: JavaScript
• Homepage: https://getlosh.xyz/dex
• Size: 4.73 MB
• Stars: 10
• Watchers: 1
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# dex

Visual Pokedex for LLMs

![image](https://github.com/user-attachments/assets/ca5d0aa8-4b04-4045-ac7a-5faf27fb4af1)

# demo

go to live demo with preprocessed models

run your own models in google colab

![gifdemodex](https://github.com/user-attachments/assets/89098710-7d42-442b-a306-8835ff8a88a2)

# setup instructions

- after this just do `npm install` and `npm run dev` in the `dex` folder. This should start the application.

- for deploying to remote server use `npm run build`

# how this works

- model repos on hf contains metadata files and the transformers repo contains the classes which are supposed to handle a model. we extract that information and save it in assets so that all of this can run in browser. we use torch device meta to avoiding loading stuff into memory.

- in case of being used from inside notebook, we do not write any files but keep the processed data in memory, and pass that information as query string to the iframe

# citation

```

@article{attentionmech2025dex,

  title={dex: interactive visual explorer for open source LLMs},

  author={attentionmech},

  year={2025}

}

```

# trivia

[This is the post](https://x.com/attentionmech/status/1907967237744468220) where it got started, and this is [where it got viral](https://x.com/attentionmech/status/1908677321688506469).