Owen L

icon location SF icon location Computing

Results 338 comments of Owen L

More documentation problems

Update: the SpinSystemInfo page changed, but is still wrong, now it just says `A namedtuple alias for field number 0` for each attribute

Having two PQC layers with the training circuits in between.

There are probably several ways to code this, but two immediately come to mind. If T is of a constant circuit structure (or is a small finite set of structures)...

Having two PQC layers with the training circuits in between.

No problem! Any updates on this issue or can it be closed?

Kernel freeze at tf.keras.Sequential.fit()

Any updates on this issue @rafalpotempa or can it be closed?

Sharding integration is much slower than pmap

Note, this is not universal that sharding is just slow on my machine due to some specific problem with my CPUs/install. Running a simpler version of the equinox example yields...

Sharding integration is much slower than pmap

This did accidentally include compile time, but I took them out and it has no real impact on times (updated the code above).

Sharding integration is much slower than pmap

I will try on GPUs and report back

Sharding integration is much slower than pmap

I slightly adjusted the code just to make some things a little faster (just reduced integration time). In addition, this is now with 4x A100s, rather than 10 (Mac) CPUs....

Sharding integration is much slower than pmap

I added the EQX_ON_ERROR, then I started to slowly roll back complexities (until we would have a manually stepped Euler ODE) when I noticed something. Stepsize controller _with_ saveat ts...

Sharding integration is much slower than pmap

I was looking more into the while loop, since we have a double while loop as you described, the first is static and the second may vary in length depending...