Its trained at native MXFP4 with MXFP8 activations layer, so you need around 1.5TB of VRAM to fully offload it without taking into account the context cache. It might be doable to do some smart expert offloading and swapping, but expect minimum 500GB of VRAM and 1TB of system RAM minimum and the t/s would be reduced.

Its only realistically runnable on datacenter grade gpu at decent speed for now (and judging from the price of ram for the next few years)

Vim/neovim if you are willing to take the time to learn how to use / config it.

Micro otherwise, very sensible keybinds and ux that map to modern gui apps, though it doesn’t come installed in most distro unlike ed and nano.

It provides a lot of nice syntactic sugar that you would otherwise have to write a wrapper for on top of Fetch. Built in request interception, request transformation, (de)serialization, shared request config, timeout/retries management, …

Though this definitely comes with bloat and supply chain risks.

Do you use features like permissions, channels, threads, bots, … a lot ?

Application using xmpp as a backend are great and can be all you need if you just want a simple voice/text chat with no additional features.

There is no equivalent to most “server/community” features of discord in the xmpp protocol, and it is something that needs to be implemented at that level for it to work.

I like xmpp, but it is not a discord alternative. It is a WhatsApp / Signal / iMessage alternative.

It doesn’t have 80% of discord features, I use discord a lot and I don’t have a single group chat.

Matrix / Element is a way better alternative feature wise.

4.0.0 was the release of the Rust rewrite

The support is bad for custom nodes and NPUs are fairly slow compared to GPUs (expect 5x to 10x longer generation time compared to 30xx+ GPUs in best case scenarios) NPUs are good at running small models efficiently, not large LLM / Image models.

Most website do not run on servo

Definitely looks worse, but I can see why they made the change, gtk is a pain to work with compared to qt