bbycroft / llm-viz Goto Github PK

3D Visualization of an GPT-style LLM

TypeScript 91.79% Python 0.39% JavaScript 0.49% SCSS 2.29% Odin 4.15% Shell 0.13% Assembly 0.54% C 0.16% CSS 0.06%

llm-viz's Introduction

Brendan Bycroft's Home Page & Projects

This repository contains my (Brendan's) homepage, as well as a number of non-trivial projects.

They are kept in a single repository for ease of deployment, as well as sharing a bunch of js utils which are otherwise a pain to share around.

Projects

The main projects are:

LLM Visualization: 3D interactive model of a GPT-style LLM network running inference.
[WIP] CPU Simulation: A 2D digital schematic editor with full a execution model, showcasing a simple RISC-V based CPU

LLM Visualization

This project displays a 3D model of a working implementation of a GPT-style network. That is, the network topology that's used in OpenAI's GPT-2, GPT-3, (and maybe GPT-4).

The first network displayed with working weights is a tiny such network, which sorts a small list of the letters A, B, and C. This is the demo example model from Andrej Karpathy's minGPT implementation.

The renderer also supports visualizing arbitrary sized networks, and works with the smaller gpt2 size, although the weights aren't downloaded (it's 100's of MBs).

CPU Simulation (WIP; not exposed yet!)

This project runs 2D schematic digital circuits, with a fully fledged editor. The intent is to add a number of walkthroughs, showing things such as:

how a simple RISC-V CPU is constructed
the constituent parts down to gate level: instruction decode, ALU, add, etc
higher level CPU ideas, like various levels of pipelining, caching, etc

Running Locally

Install dependencies: yarn
Start the dev server: yarn dev

llm-viz's People

Contributors

Stargazers

Watchers

Forkers

kustomzone shaneholloman simpzan codeaudit pkiage luvwinnie fukcinglife binaryninja carsontang trocker hbcbh1999 ryman warkanlock quantza zzkcaesar jeffara kemolo eramax mchorfa dacapoday ghosthamlet haz nb-xx davidgarciamontreal to3d pomelo-nwu yifree gptalgopro kimi-gao jolks megamis zcfrank1st guoqiyi git-models christiansch thmosxiong xstarlink shadown tofun666 apollohuang1 jshekhawat hgslv rkp64 tomputer22 dazzaji iblech projstudy aihuasxy mrku69 holm-xie carlosgdcj paybaxanekowolfie scomants-0 56-gambitleak letsuchem-l thesoyithugzhoff 63-gorgeousguru edson-github chattypyrei 12isoleph richardsonjf loicnestler axwerbit-y mundefr-fource messagesta-backinfo mahopenta-mercyfeline bozsyineedmoon nicsyscalamarket finmalage-westphold choneprep22chikdot slipkray-z qbloodsol bloggeno14 beathite16 whippeddomeaming sevarica30 portestu48 p-centhart lilllepiclife skyck-skunkylysi cuecardsh81 ilovecrypto17 timholds tutumomo hhy5277 devenlu cigncloud kanehui mikelmanro torobirgitte amitobe-code lenalanini vaaaas yanxg xushaohui reallyarshad tuhinmallick eltociear using-namespace-system mbrukman

llm-viz's Issues

Feature Request : BERT Visualisation

Thanks for this amazing work !

Is there a pipeline on generating this visualisation for new models ? I would love to get something running on BERT or T5 architecture. How can I achieve this ?

No such file named "BUILD_ID"

Thanks for your nice work.
When I was running this project, I encountered the following error after running "npm run start":

D:\Downloaded\llm-viz>npm install
npm WARN deprecated [email protected]: Package no longer supported. Contact Support at https://www.npmjs.com/support for more info.

added 545 packages, and audited 546 packages in 3m

161 packages are looking for funding
  run `npm fund` for details

3 vulnerabilities (1 low, 2 moderate)

To address all issues, run:
  npm audit fix --force

Run `npm audit` for details.

D:\Downloaded\llm-viz>npm run start

> start
> next start

- ready started server on [::]:3000, url: http://localhost:3000
[Error: ENOENT: no such file or directory, open 'D:\Downloaded\llm-viz\.next\BUILD_ID'] {
  errno: -4058,
  code: 'ENOENT',
  syscall: 'open',
  path: 'D:\\Downloaded\\llm-viz\\.next\\BUILD_ID'
}

What should I do besides running npm install, before running npm run start?

RWKV

I love the visualisation in https://bbycroft.net/llm.
It would be great to see how RWKV works.

Do you accept translation?

Hi.
Thanks for your amazing job

Do you have any place for accepting the translation contributions?

how can i use the code?

I have git clone this project, and install yarn。but how can i run this code?

Typo in "Chapter: Layer Norm"

I think there is a typo in the second step.

We can regard each column separately, so let's focus on the 4rd column (t = 3) for now.

Should be 4th.

GPT-2 Viz

Great project! It helps me easier to understand how gpt works.

Could I know how to viz the gpt2 network. Currently, it defaults to nano-gpt.

I just want to confirm if output is correct or not

as mentioned, Its goal is a simple one: take a sequence of six letters:
C B A B B C

and sort them in alphabetical order, i.e. to "ABBBCC". but output is CBAAAA

the last part maybe can have more of a "goodbye" thing, I did not understand it was done

maybe like "tjanks folks that's all", or maybe a CTA idk, maybe the last part the very last part can have some animation which is perhaps missing and maybe a final result thing, idk, just ideas, I loved the whole thing, it motivated me to read the code. thank you for building this, you are awesome

nano-GPT train for letter sorting

Great project! I would like to know how to train the letter sorting model mentioned in the nano-GPT visualization project? I think the visualization here is more clear when combined with the nano-GPT project code? So is it convenient to provide training documents for the alphabetical sorting model? Thank you so much!

Cheers,
Dan