-
I'm not sure about this but with the modification made yesterday in the loss, If we have a batch full of images without annotations, the "regression part" does not get the .backward() call which may l…
-
With efficient `fvp` and `ggnvp` #63 it might be nice to add the Hessian-free optimization algorithm of https://www.cs.toronto.edu/~jmartens/docs/Deep_HessianFree.pdf
This algorithm runs a short co…
-
### Describe the feature or idea you want to propose
Our new experimental forecasting module starts with a really fast ETS implementation. The next stage is AutoETS. But how to search the parameter s…
-
Gradient check for an `AmiciObjective`
```python
pypesto_problem.objective.check_gradients_match_finite_differences(
x=startpoints[0],
)
```
invokes an error:
```
Traceback (most r…
-
You set an absolute tolerance of `g_tol = 1e-8` by default. I don't know how you picked this — it depends on the scaling of the problem, because **gradients are dimensionful**.
This implicitly as…
-
i use this set
# --gradient_checkpointing \
--mixed_precision fp16 \
--use_8bit_adam \
--set_grads_to_none \
but error is
return Variable._execution_engine.run_backward( # Calls i…
-
Using my latest [CCV-98720.xml.zip](https://github.com/user-attachments/files/17550675/CCV-98720.xml.zip), I noticed serious bugs. Eg:
~~~
J163->J17: Shimming Toolbox: An open-source software tool…
-
First of all, thanks to the authors for their excellent contribution to the diffusion modeling community, sid is definitely the coolest distillation method I've ever tried (and more interesting compar…
-
```
The Grad menu only allows the user to choose a "start" and "end" color, and
the gradient comprises of all the colors in-between.
It would be better to allow select colors manually.
- Requires mod…
-
```
The Grad menu only allows the user to choose a "start" and "end" color, and
the gradient comprises of all the colors in-between.
It would be better to allow select colors manually.
- Requires mod…