Hey, do you have same info from pytorch users?
-
-
-
No, these were mostly Tensorflow users and some had used Pytorch occasionally. I didn't ask for a comparison between Pytorch and jax. My sample is biased towards Google/Deepmind researchers, for whom TF1, TF2, jax are the only feasible choices.
- Još 3 druga odgovora
Novi razgovor -
-
-
I really wanted to like JAX, but didn't. It seems awesome for "core DL" research, but feels underdeveloped for the use-case where DL is primarily a tool in service to a different direction (in my case, RL). Needs more tooling.
-
Also I think
@carlesgelada had some real grievances with the way that random numbers are implemented by JAX...having to manually query & pass a seed generator is pretty obnoxious. - Još 13 drugih odgovora
Novi razgovor -
-
-
i've done my last couple of mini prjs in TF2.0, pytorch & jax. there is a lot of nice stuff around the core of jax but there's still such a huge amount of momentum re: tooling & infra from TF that means i'll stick with it for awhile yet. ( note is said: TF2, _not_ TF1!, eagerFTW)
- Još 3 druga odgovora
Novi razgovor -
-
-
It’s pretty cool to see projects like rewriting David Mackey’s Bayesian NN examples in JAX. Suggests it has a lot of energy behind it.
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
What about compared to pytorch?
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
This reminds me of this tweet (although for Pytorch):https://twitter.com/karpathy/status/868178954032513024?s=20 …
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
jax is a thang now
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.