Support differentiability through Torch tensors #703
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Typing is still failing
* works with basic features of pulser-diff * Fixed phase attribute setting; removed debugging code; reverted unnecessary changes * Modified register creation code to work with AbstractArray; register coordinates are differentiable with pulser-diff * Fixed type hints * Minor fixes and refactoring * Modified ParamObj code to work with quantum model training in pulser-diff * Minor refactoring; add possibility to ensure 0D AbstractArray is reshaped into 1D * Force array only for scalars
a-corni
reviewed
Aug 14, 2024
a-corni
reviewed
Aug 14, 2024
a-corni
reviewed
Aug 16, 2024
a-corni
reviewed
Aug 16, 2024
a-corni
reviewed
Sep 12, 2024
Merged
HGSilveri
added a commit
that referenced
this pull request
Sep 20, 2024
**Main changes:** - Reworking the NoiseModel interface (#710) - Allow modification of the EOM setpoint without disabling EOM mode (#708) - Enable definition of effective noise operators in all basis (#716) - Add leakage (#720) - Support differentiability through Torch tensors (#703) - Add from_abstract_repr to Device and VirtualDevice (#727) - [FEAT] Handle batches with partial results (#707) - Add open batches to pulser-pasqal (#701)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR adds native support for
torch.Tensoron select quantities, so that the final Hamiltonian can be differentiated wrt to these parameters. A few important design considerations were taken here:torchwill be an optional dependency: Users that don't require this feature should be able to keep installing and using pulser as before. For those who do want thetorchsupport, they can have it by adding[torch]to theirpip installcommands:# Both options work pip install pulser[torch] pip install pulser-core[torch]np.ndarrays andtorch.Tensors indiscriminately, the wrapper classAbstractArrayis created. This class contains an array-like object, which is stored either astorch.Tensoror anp.ndarray. To the largest extent possible, instances of this class can be handled like array, allowing operations, indexing, etc... However, their use should be only internal, so they contain only the necessary functionality for the existent codebase - whenever extra functionality is needed, it can always be added.AbstractArraymust use custom functions to preserve theAbstractArraytype in the output. As such, analgous definitions of the necessarynumpy/torchfunctions were defined in the newpulser.mathmodule. This new module is, by convention, imported withimport pulser.math as pmand can often be used as a drop-in replacement fornp.differentiabilitytensor-like behaviour are marked with the newpm.Differentiablepm.TensorLiketype-hint. When this is present, atorch.Tensor(withrequires_grad=Trueor not) can be provided. Whenever a parameter is not marked aspm.TensorLike, providing atorch.tensoris not guaranteed to work as intended.