Hacker Read

HanClinto · 2023-09-27 20:44:39

The code I linked is the model source -- the structural skeleton. You can create a model with this source and put whatever weights you want in it -- just like Doom.

spywaregorilla | karma 6714 | avg karma 2.33 · | 2023-03-12 11:32:02

The weights are the model.

The small bits of standardish code to utilize them are not that material

reply

catgary | karma 407 | avg karma 2.5 · | 2024-06-01 16:17:16

The weights are like jpegs, the model itself is still a piece of software that executes code.

exizt88 | karma 2218 | avg karma 15.19 · | 2023-07-06 04:30:08

The model code is Apache 2.0, the weights are proprietary.

jerpint | karma 901 | avg karma 1.97 · | 2023-04-16 06:11:00

They actually reimplemented the code to load the model. But it’s not clear if the weights of the model are free to use or not

swyx | karma 19447 | avg karma 4.18 · | 2023-06-14 20:11:59

the code is open source but not the weights. as far as i can tell.

theaniketmaurya | karma 4 | avg karma 0.2 · | 2023-03-28 12:40:24

I am curious though how would the model weights work out?

rytill | karma 681 | avg karma 1.83 · | 2022-12-22 14:46:18

Are there model weights?

spencerchubb | karma 1140 | avg karma 2.74 · | 2024-02-08 21:02:33

That is an awesome idea. I wish the authors would open source the code and weights so this can be tried.

navjack27 | karma 707 | avg karma 0.64 · | 2022-09-13 04:33:14

What do you think it is? Some corporeal being? It's an overgrown python script pulling weights out of a model.

emadm | karma 401 | avg karma 2.88 · | 2024-02-13 19:14:48

Code is MIT, weights are under the NC license for now.

nerdponx | karma 22397 | avg karma 2.51 · | 2023-07-05 11:25:37

Weights are equivalent to compiled object code IMO. All else follows from there.

nerdchum | karma 651 | avg karma 2.03 · | 2023-04-03 15:48:47

what are model weights?

simonw | karma 58201 | avg karma 7.31 · | 2023-07-18 15:40:18

When you run that command, where does it download the model weights from?

ewuhic | karma 307 | avg karma 2.65 · | 2022-05-15 16:11:01

Isn't the "why" the weights of the model themselves?

tvirosi | karma 117 | avg karma 2.34 · | 2021-08-18 09:22:03

Probably the model weight files

nmfisher | karma 3238 | avg karma 3.02 · | 2023-08-28 03:03:19

No that's what I was getting at. For roughly half the models I come across, the weights are published on HF. For the other half, the weights are available via GitHub (either the releases section, or a link to something like Google Drive).

2OEH8eoCRo0 | karma 10664 | avg karma 1.95 · | 2023-03-24 08:12:37

Model weights seem akin to publishing a decompiled binary.

carefree0910 | karma 135 | avg karma 5.19 · | 2022-10-30 11:57:15

I noticed that they added a `LICENSE-ModelWeights.txt`! Maybe I should dive into their repo deeper and follow their good practices.

Thanks a lot for the advice! It really helps!

reply

Hendrikto | karma 2515 | avg karma 3.67 · | 2023-02-22 06:02:14

It is definitely possible. At any point, you can just take a snapshot of the weights. Together with a description of the architecture, this is a complete description of a model.