-
## 🐛 Bug
The gumbel_softmax function returns seemingly random output when inputs are distributed around 0 to 1. I think this problem arises from the Gaussian noise added to the softmax; if inputs a…
-
### 请提出你的问题 Please ask your question
使用设备为macbook pro M2 ，系统版本14.4.1
在使用pip安装的paddlepaddle（包括2.3.2，2.4.2，2.5.2），均在运行静态图的infer过程中hang住，其中2.3.2还会出现segmentFault后退出
采用的是官方教程中的示例代码，包括使用inference API…
-
-
Thanks for your nice work.
I am a little confused about training a VAE model without sampling latent variable z mentioned in paper:
> Notably, the Gumbel-Softmax reparameterization is not used to b…
-
### 🚀 The feature, motivation and pitch
We are working on a research project, where we need to prevent back propagation between some nodes while allowing for forward propagation. This represents a gr…
-
When i was trying to train with multiple GPUs due to the limited memory, the loss was not stable and can not get the ideal results. But when i using 1 GPU to train, the loss is less than 0.1 and the t…
-
_placeholder for brainstorm_. Finished all master courses. (part-time side job)
Exploring for 1 month what a good master thesis direction is around LLM.
Draft master thesis (again placeholder): *…
-
Could you please explain why certain datasets require a threshold and why there are different thresholds for them during inference?
` if self.training:
# During training
…
-
### 🐛 Describe the bug
NaNs are sometimes in the output when running on CPU. I did not find the exact pattern, but when sampling from the same tensor long enough nan will appear.
```
import torch
…
-
## 🐛 Bug
'torch.nn.function.gumbel_softmax' yields NaNs on CUDA device (but not on CPU). Default parameters are used (tau=1, hard=False).
## To Reproduce
The following code generate rando…