←back to thread

899 points georgehill | 8 comments | | HN request time: 0.823s | source | bottom
1. yukIttEft ◴[] No.36216376[source]
Its graph execution is still full of busyloops, e.g.:

https://github.com/ggerganov/llama.cpp/blob/44f906e8537fcec9...

I wonder how much more efficient it would be when Taskflow lib was used instead, or even inteltbb.

replies(4): >>36217006 #>>36217540 #>>36217840 #>>36218226 #
2. moffkalast ◴[] No.36217006[source]
Someone ought to be along with a PR eventually.
3. boywitharupee ◴[] No.36217540[source]
is graph execution used for training only or inference also?
replies(1): >>36217851 #
4. make3 ◴[] No.36217840[source]
does tbb work with apple Silicon?
replies(1): >>36217968 #
5. LoganDark ◴[] No.36217851[source]
Inference. It's a big bottleneck for RWKV.cpp, second only to the matrix multiplies.
6. yukIttEft ◴[] No.36217968[source]
I guess https://formulae.brew.sh/formula/tbb
7. mhh__ ◴[] No.36218226[source]
It's not a very good library IMO.
replies(1): >>36224140 #
8. dindresto ◴[] No.36224140[source]
ggml or Intel TBB?