Qwen3 availability

[mfleucha]

Hi,

not sure if I am doing something wrong but I'm wondering if Qwen3 could be supported soon? Should I already be able to use it (deployed latest version)? If yes, I'd appreciate some help finding the way as I didn't when setting up and checking the docs...

Thanks a lot for a supercool piece of software!

mfleucha

[sfavorite]

Looks like this project has been abandoned. There is a fork you can checkout.

https://github.com/digitalmint/exo

[ro-mak]

this guy's PR was closed and according to his repo https://github.com/shamantechnology/xotorch

Exo v1 is trully abandoned because Exo v2 will be close source

you can try though to add new models to models.py

as far as I understand you should check tinygrad or mlx supported models

if you want to use macOS you have more options available via mlx

if you have other system you have to wait until tinygrad supports something then update dependencies and add new model to models.py i guess

I am not a contributor, just a guy who managed to connect two windows notebooks with WSL and a macbook into a cluster) and found out that tinygrad only supports Llama and even those are hard to load because they are either too big or too small

[Nurb4000]

For what its worth llama.cpp does support distributed machines. May be an outlet since this project is dead.

[requiredtruth]

@ro-mak @sfavorite Your a freaking life-savers, down to hour. Good work informing us, Thank you very much

[water-vapor]

you can try my fork: https://github.com/water-vapor/exo
I added qwen3 and qwen3_moe support

[fakebizprez]

this guy's PR was closed and according to his repo https://github.com/shamantechnology/xotorch

Exo v1 is trully abandoned because Exo v2 will be close source

you can try though to add new models to models.py

as far as I understand you should check tinygrad or mlx supported models

if you want to use macOS you have more options available via mlx

if you have other system you have to wait until tinygrad supports something then update dependencies and add new model to models.py i guess

I am not a contributor, just a guy who managed to connect two windows notebooks with WSL and a macbook into a cluster) and found out that tinygrad only supports Llama and even those are hard to load because they are either too big or too small

is this for sure? they're pulling the rug already?

[Evanev7]

We have qwen3 support in 1.0!