LoganDark
LoganDark
@ggerganov what do we have to do to get this PR merged? Is the CI failure important? -Emily
> Nothing on your side, I'll merge this later today So how's today going so far? lol
> * support multi-thread operators > * pass `ggml_compute_params` to the callback Indeed these are things that I considered, but my impression was that `ggml_compute_params` was an unstable internal implementation...
> they are subset of the new custom ops Not really, those are for pure functions that can be parallelized by row. Useful to keep around while the user can't...
imagine running ggml on CHERI though, which will hardware fault if you try to use a mere integer as a pointer
> > imagine running ggml on CHERI though, which will hardware fault if you try to use a mere integer as a pointer > > Do you mean [CheriBSD](https://www.cheribsd.org)? No,...
That's actually hilarious that I correctly pointed out CHERI if that was the real motivation for this pull request :D
Hm, this can be done manually by saving and resetting the `n_nodes` and `n_leafs` fields of `ggml_cgraph`. You can chop off trailing computations that aren't needed. It actually seems to...
> Why not just write the entire thing in one Mojo file. Because Mojo is vaporware and doesn't actually exist yet? You can't yet download or run Mojo, it is...
> There's a reason they're close-sourcing it for now, the same reason I close-sourced MFA for ~2 months. It's too buggy at the moment and will develop faster the way...