Arthur Lagacherie

17y/o 🇫🇷 | LLMs are all you need
Training models, breaking things, sharing what works.

[Arthur Lagacherie @arthurlge]

@Blesk_FR hey, im working on low bits quantization for llms, i would love to connect

[Arthur Lagacherie @arthurlge]

@opencode not available in france 😑

[Arthur Lagacherie @arthurlge]

@opencode a french model !!

[Arthur Lagacherie @arthurlge]

@runcrate I'm currently running on very small model (56m params) but I'm running it on my mac so it's slow

[Arthur Lagacherie @arthurlge]

I really need GPUs, each experiment I'm running take forever!

[Arthur Lagacherie @arthurlge]

@Perce_gd hi, im into llm quantization, lets connect 👍

[Arthur Lagacherie @arthurlge]

the new one is quite beautiful 😁

[Arthur Lagacherie @arthurlge]

going straight with his brothers

[Arthur Lagacherie @arthurlge]

@kendrickkumarr @X hey, im working on sub-1-bit llm quantization, i would love to connect 👍

[Arthur Lagacherie @arthurlge]

@RajRitwik74853 hey, im 17 too and im working on llm quantization!, lets connect

[Arthur Lagacherie @arthurlge]

@guptankit27 hey, im working on llm quantization, lets connect 👋

[Arthur Lagacherie @arthurlge]

@princekush181 to achieve better performance on my sub-1-bit quantization method to scale it on bigger models

[Arthur Lagacherie @arthurlge]

Today I've made great improvements to my sub-1-bit quantization. It might be scalable soon.

I also started reading a book about kernel methods to better understand the math. Who knows, maybe it'll be useful for quantization or other projects.

[Arthur Lagacherie @arthurlge]

@anurudhsharmaa hey, i'm obsessed with LLMs, would love to connect 👍

[Arthur Lagacherie @arthurlge]

@pujulieeeeeeee pretty simple
llm->weights->shrunk->train->end

(and a few technical details without which it wouldn't work 😂)

[Arthur Lagacherie @arthurlge]

i've been running experiments on @askalphaxiv's OpenResearch for the past 3 days, and I currently have 60+ leaves in my tree, each representing an idea.

i really recommend OpenResearch if you're working with agents to run experiments

[Arthur Lagacherie @arthurlge]

study sub-1-bit quantization of LLMs

it's all you need

[Shriya🌻 @Shriyaaa_06]

How to become addicted to studying?

[Arthur Lagacherie @arthurlge]

@dcdigit hey, i would love to connect, just followed you 👍

[Arthur Lagacherie @arthurlge]

@RoliumGens @MiaAI_lab oep, mais je savais pas que grok restreignait en france aussi