Skip to content

Open source Claude Artifacts – built with Llama 3.1 405B

Notifications You must be signed in to change notification settings

taoztw/llamacoder

This branch is 206 commits behind Nutlope/llamacoder:main.

Folders and files

NameName
Last commit message
Last commit date

Latest commit

003ce16 · Aug 10, 2024

History

57 Commits
Aug 10, 2024
Aug 7, 2024
Aug 2, 2024
Aug 8, 2024
Aug 8, 2024
Aug 9, 2024
Aug 9, 2024
Aug 2, 2024
Aug 2, 2024
Aug 2, 2024
Aug 2, 2024
Aug 9, 2024
Aug 2, 2024
Aug 9, 2024
Aug 9, 2024
Aug 2, 2024
Aug 2, 2024
Aug 2, 2024

Repository files navigation

Llama Coder

An open source Claude Artifacts – generate small apps with one prompt. Powered by Llama 3 405B & Together.ai.

Tech stack

  • Llama 3.1 405B from Meta for the LLM
  • Together AI for LLM inference
  • Sandpack for the code sandbox
  • Next.js app router with Tailwind
  • Helicone for observability
  • Plausible for website analytics

Cloning & running

  1. Clone the repo: git clone https://github.com/Nutlope/llamacoder
  2. Create a .env file and add your Together AI API key: TOGETHER_API_KEY=
  3. Run npm install and npm run dev to install dependencies and run locally

Future Tasks

  • Add dynamic OG images to the specific generations & include the prompt
  • Add more dynamic OG images for playwright
  • Address issue of ability to publish the same app repeatedly
  • Try chain of thought reasoning to see if it works better overall
  • Encourage best practices by making the input and textarea & having pills to generate apps w/ good prompts
  • Save previous versions so people can go back and forth between the generated ones
  • Could be nice to show a "featured apps" route on the site on /featured. Have a /id/${prompt} dynamic route that can display a bunch of nice example apps in the sandbox ready to go
  • Support more languages starting with Python, check out E2B
  • Try finetuning a smaller model on good prompts from 405b or GPT-4/Claude
  • Add dark mode to the site overall, nice design change
  • Think about how to have 405B correct itself (sometimes it makes up imports)
  • New route for updateCode that only sends the latest generated code + the modify request
  • Fix bug where if a user edits the code, then does a change, it doesn't use the edited code
  • Add rate limiting with redis upstash if the traffic gets too high
  • Try to add a consistent component library like shadcn
  • Apply code diffs directly instead of asking the model to generate the code from scratch
  • Add the ability to upload things like a screenshot for it to start from that

About

Open source Claude Artifacts – built with Llama 3.1 405B

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • TypeScript 95.7%
  • CSS 3.7%
  • JavaScript 0.6%