Add qjit-ted "Symmetry-invariant quantum machine learning force fields" demo #1392
Conversation
|
To me here are the possible paths forward: (a) make lightning work with differentiating through qubit unitary (will take a bit of time) Personally, (c) makes the most sense for me at this stage. |
| num_batches = 200 # 5000 # number of optimization steps | ||
| batch_size = 5 # 256 # number of training data per batch | ||
|
|
There was a problem hiding this comment.
This smaller-sized run can finish in a reasonable amount of time (around 5 minutes)
|
Hello. You may have forgotten to update the changelog!
|
|
@paul0403 I recommend not adding the demo here -- I am in the process of moving all demos out of the Catalyst docs and into pennylane.ai/qml, to ensure that:
|
I was not aware of this! Thanks for letting me know! |
Context:
As part of Catalyst's work on identifying 10 demos to compile with catalyst, we convert "Symmetry-invariant quantum machine learning force fields", a machine learning workflow that calls a training step function repeatedly and thus can have significant performance boosts with catalyst compilation.
Description of the Change:
Convert the demo to use qjit instead of jax.jit
Benefits:
Demonstration of qjit advantage
Possible Drawbacks:
Because lightning does not support differentiation through
QubitUnitary, we have to resort to finite difference method for gradients, which brings a big performance degradation.Related GitHub Issues:
[sc-72938]