Alex Cheema
Alex Cheema
This looks good, however this is the "easy" part. The tricky part is packaging exo as an installable in the first place and ensuring it works across all platforms. I'll...
I think I know what's going on here. The assumption we make on this line is not true: https://github.com/exo-explore/exo/blob/2654f290c3179aa143960e336e8985a8b6f6b72b/exo/networking/udp/udp_discovery.py#L143 i.e. `self.known_peers[peer_id][0].addr()` is not always of the form `{peer_host}:{peer_port}`. This would...
Can you try again? I hope #278 fixes this for you.
The requirement is there because exo uses asyncio, which has bugs on older versions of Python. Which devices can't run Python>=3.12? On another note, I've been working on a rust...
This is a great start - I tested this and it works. That's awesome because that means we can support any mlx model in tinygrad. Are you sending parameters in...
This looks great. What we'd really need here to get this merged is automatically sharding the model. The PyTorch implementation here does this automatically, you can see how they do...
Ideally we do this like #139 where we support models automatically without needing separate `build_{model}` functions.
Hi Vijay, this would be a great addition - we need better support for CPU in exo! Did you forget to commit some files?
What OS are you running on?
> Hi @AlexCheema can you give me some more insight into what you want to be refactored? I initially thought we could reuse some of the download percentage logic that...