atamyrat

Iň köne account 🦕

[atamyrat @atamyrat]

PlayStation Portable https://t.co/CsLo3JfgZB

[atamyrat @atamyrat]

Enjoying all these llama2.c ports running on exotic platforms: Amiga, Atari, Xbox, PS3, PSP, Windows 98, Solaris/Sparc, Sega.. I wanted to do Mega Drive/Genesis one but was surprised it did not have floating point unit!

[Alex Cheema @alexocheema]

"Somebody got one of the small versions of Llama to run on Windows 98...We could've been talking to our computers in English for the last 30 years" - @pmarca

It was me! I got Llama running on a Pentium II machine with 128MB RAM running Windows 98. Details below.

[atamyrat @atamyrat]

Nintendo 3DS

[Danyal @_danyalmalik]

Not really much of a "port" but I recently ran @karpathy's llama2.c on my old 3ds. Crazy to see a language model running on that thing lol
https://t.co/A8Ey327ULs

[atamyrat @atamyrat]

@mariotelfig @francoisfleuret karpathy/llm.c

[atamyrat @atamyrat]

@francoisfleuret @mariotelfig karpathy/llama2.c

[atamyrat @atamyrat]

@francoisfleuret I was thinking as a weighted regularization term, implemented with torch.distributions.categorical entropy(). Haven’t tested yet though, for batched preds it might be easier to write the formula directly.

[atamyrat @atamyrat]

I’ll take the blame for a line or two here 🙋‍♂️

[Jim Keller @jimkxa]

Andrej, @karpathy, I think this is the marketing MatMul https://t.co/we6sV0p868
// do the matmul in groups of GS
int j;
for (j = 0; j <= n - GS; j += GS) {
for (int k = 0; k < GS; k++) {
ival += ((int32_t) x->q[j + k]) * ((int32_t) w->q[in + j + k]);
}
val += ((float) ival) * w->s[(in + j) / GS] * x->s[j / GS];
ival = 0;
}
I wish it worked this way

[Yann LeCun @ylecun]

That's pretty insane!

[Max Braun @maxbraun]

Hack of the day: Llama on a microcontroller

🦙🔬

Details and source code at https://t.co/CHzkyqIArh

Thanks to @karpathy, whose llama2.c inspired this!

[atamyrat @atamyrat]

🦙 @karpathy’s llama2.c is an exciting project. Happy to see my contribution accepted! https://t.co/XAkD2R2LlS