-
Xploder mentioned it on discord that state, brightes and so on are always in the payload. It would be better to send only the changed properties to the lamp.
This implies that we should also remove…
-
When training batch size 4 on H100 the speed is 1.27 second / it
When training batch size 4 on 2x H100 the speed is 2.05 second / it
So basically we almost got no speed boost from multiple GPU t…
-
The training loss is chosen to be Huber or L2. Why not use LPIPS loss as was done in the original consistency models paper after decoding the unet latents via the VAE? It showed better FID.
I think…
-
It seems that Rrunfile.m should be renamed to Rread.m to avoid this error:
```
Unrecognized function or variable 'Rread'.
Error in test_huber_smsp (line 12)
Rread('Rscript.R')
```
-
Maybe write short one liners that tell users what things like berhu, huber, quadform, hmax, hmin do - they are difficult to understand like that. Giving a link to a wikipedia article for further expla…
-
RMSE and MAPE have their issues in certain contexts, so I just like to put in a request for additional error measures such as MASE, MAE, Huber, etc., or maybe a function or argument that allows the us…
ercbk updated
4 years ago
-
We now save models using torch.save() and torch.load(). This is the recommended way for pytorch as far as I could find: https://pytorch.org/tutorials/beginner/saving_loading_models.html#:~:text=Saving…
-
1. Open https://bakery-flow.demo.vaadin.com/storefront
2. Click the first item (Lionel Huber)
3. Click "edit"
Edit dialog is shown, not sure if it's another popup or the same
4. Click "cancel"…
-
### 🚀 The feature, motivation and pitch
Port the following operator to TensorIterator. This should improve performance and make the code cleaner, see the links below for more context.
https://pyto…
-
dislaimer: this is a nitpicking issue.
in my experience, the include of pdf files in latex documents produced more clear and crisp images than using png/jpg as saved by matplotlib. Maybe @dafnevk o…