Skip to content

Conversation

@subangstrom
Copy link
Collaborator

No description provided.

hexane360 and others added 20 commits September 5, 2025 15:16
* Start of torch backend

* More torch utils

* Most torch utils working

* Got ePIE working on torch

* Renamed cupy backend, generalizing dataclass

* Typing fixes

* More torch work, basic grads running

* Gradient descent solver working

* More solvers with torch

* Add back in dtype checks

* Implement affine_transform order=0,1 on torch

* More device selection abstractions

* torch fixes and improvements

* Add torch optional dependency
* Fix drop_nan_patterns with tilt

* initial setup

* dm4 data at least loading, may need flips, etc

* add flips after gatan read-in

will need to move if in metadat

* turned off memmap on loading for improved performance

* fixed flipped (done by reader)

* adding an offset hook

* fixed an error

* including a transpose

* need to keep the mask

* issue fix with scan shape

* reading in the e- calibrations and appying to patterns

* line cleanup

* cleanup

---------

Co-authored-by: jmlebeau@gmail.com <jmlebeau@gmail.com>
Co-authored-by: subangstrom <lebeau@mit.edu>
* Add sign option to tree.grad()

* Report granular/detailed losses

Need to update conventional engines, as well as server/web interface

* Granular losses for conventional engines

* Fix web interface w/ granular losses

* Typing fixes

* Print detailed error breakdown

* Hotfix for prepare_for_engine backend error
@hexane360
Copy link
Owner

Can you make a PR to the develop branch instead of main?

@hexane360 hexane360 closed this Nov 3, 2025
@subangstrom
Copy link
Collaborator Author

subangstrom commented Nov 3, 2025 via email

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants