Stavros Kassinos

Ok, great thank you for the quick response 🙂 I will use the cpu workaround, just to get a placeholder function ready in my code. BTW, for physics-based problems having...

I am getting the exact same error every time. Does not work.

I second that. einops saved me quite a few times where the logic was becoming a bit intractable.

I can see the performance motive, but if by mistake you use an out of bounds index to set the value (i.e. you think you are setting the last value...

I am testing the performance of various schemes on MLX. For context, I am using the toy problem of solving the 2D Heat Conduction equation. Actually, one can solve the...

Ok knowing that, I will try to do some debugging and see if I can provide some additional feedback.

It runs orders of magnitude faster (438 sec vs 2.37 sec for 100K steps) !!! And whatever the bug is, it's not surfacing any more. I'm thrilled with the speedup...

I was too quick to respond. The T_old temperature field (see the code below) needs to be updated in each time step, that's why I had the line `T_old =...

Following your remark, I tested again `T_old = mx.broadcast_to(T,shape=T.shape)` and `T_old=T` and they are about the same (within load noise). I must have been misled by some other change I...

Ok, I did some basic debugging using the debugger in PyCharm (PyDev). I inserted a breakpoint at the iteration step that I would normally need to execute `mx.aval(T)` to avoid...