-
Does MadNLP currently support using a lazy Jacobian and Jacobian transpose? It should be possible to use an iterative solver to solve the linear system which doesn't require forming the Jacobian matri…
-
I updated to the most recent version of lightkurve and the most recent version of Astropy, to avoid an annoying Astropy deprecation warning. During those upgrades, I believe a version of numpy got upd…
-
I really like the idea of making differentiating part of the Multi-Output interface. With this in mind some thoughts I had on adding differentiation to KernelFunctions:
- I think until `AbstractDi…
-
There is still some slack in GLM, especially fit_gradient, to improve performance in terms of speed and memory consumption
related issues:
- #4622. PR #4624 redundant/unused pinv_exog, will be …
-
Currently, out-of-place Jacobians and Hessians use `stack` to put together the batches of JVPs/VJPs/HVPs, and there is a possible resizing at the end.
- Advantages: preserves the native types of ma…
-
Below is a list of timings I obtained for the testsuite on my desktop machine, sorted by worst to best, and everything that requires less than 45s cut out.
```
241.28 cgal/cgal_remesh_surface.debu…
-
Hi Enzyme Team,
I'm trying to compute a hessian (or a hessian-vector product) using Forward-over-Reverse, doing my best to follow what's in the documentation, but I'm crashing and can't tell if I'v…
-
There are several experiments being done with this repo to understand and evaluate the effects of quantization on the `llama2.c` models.
It is a great test-bed to analyze the effects of varying app…
-
I recognize that Vandenberghe suggests a similar (1+alpha) shift in the initialization when alpha > 0, but the '1' is a magic number that, it seems to me, should scale with an appropriate norm of the …
-
* `nP` --> `size`?
* W --> weights?
* mapPair? do we need it? Can we just duck type?
* `deriv`/2 --> gradient? hessian? Not sure on this