-
## Summary
目标是解决一个高精度的模拟计算任务,用神经网络拟合复杂方程的解。
代码部分的工作包括:1.oneflow重写训练代码;2.oneflow实现LBFGS优化算法,对标torch.optim中的lbfgs优化算法。
训练任务是相同数据训练一个10层左右的Linear网络。
目前出现问题是,同样的checkpoint,cpu版本的oneflow目前已调试到精度为99…
-
I am trying to solve a problem where `solver.run` is called multiple times to minimize a series of functions while varying a parameter. Using `memory_profiler` I can see that the allocated memory incr…
-
I've been getting NAN output with vgg16 ~50% of the time resulting in a black or scrambled-tv-reception output.
With vgg19 and NIN it happens less than 0.1% of the time.
I'm using all the same p…
-
I am trying to apply the same idea that you did but using a different dataset but running into the following error:
`ValueError Traceback (most recent call last)
[](…
-
Hello,
I'm working on a multi-classification model and the result looks good. But when I train the model after using model.prune(), train_loss and test_loss become NaN easily even if I set lr and ste…
-
Hello I
-
I tried porting the code to PyTorch, specifically the Tensorflow version. The latter works perfectly.
But with PyTorch when I use optim.LBFGS, I run into exploding gradients/no updates on the target.…
-
## 🚀 Feature
I would like to suggest new stochastic optimizer additions to Pytorch.
### For non-convex loss functions
It is known that adaptive stochastic optimizers like Adam, Adagrad, RMSprop c…
-
Hi, first of all this package looks very promising. I'm working on a problem with 70millions rows and 45million sparse features (up to ~1000 non-zero elements). I tried to make it work with Spark ML b…
-
First of all, good work, the project seems well maintained and clean : )
I have an issue around Example 4 - Classification, on the jupyter notebooks. When i try to run after following the instructi…