llama.cpp icon indicating copy to clipboard operation
llama.cpp copied to clipboard

What models i really need?

Open paulocoutinhox opened this issue 1 year ago • 3 comments

Hi,

What models i really need?

I have these:

image

The only 7B folder for example is necessary? Each model has different results?

I don't understand if i need only one and execute the training for each folder or if only one is necessary and i need choose one.

Thanks.

paulocoutinhox avatar Mar 15 '23 06:03 paulocoutinhox

Bigger model - generally better results, but uses more ram and slower. If you want fast results - 7B, best results - use whatever fits in your RAM.

terafo avatar Mar 15 '23 07:03 terafo

I have a lot of mobile projects, that is my focus, example: https://github.com/xplpc/xplpc

My github: https://github.com/paulocoutinhox

So "im enter in this world" to bring it to mobile, but im learning too about this project.

Do you think that if i generated the final file from my macbook (example: ggml-model-q4_0.bin) it works on Android/iOS or it will require a lot of RAM?

paulocoutinhox avatar Mar 15 '23 08:03 paulocoutinhox

7B is achievable on many phones (takes ~4gb of ram) I wouldn't say anything else is. So if that is your main focus, I guess ignore the rest of them. Alternatively run a server with whatever on it and have apps send the text to the server and server respond (but that uses internet/data then)

There are already some other people whom have posted videos / gits of it running on their phones.

G2G2G2G avatar Mar 15 '23 08:03 G2G2G2G