The operator 'aten::lerp.Scalar_out' is not currently supported on the DML backend and will fall back to run on the CPU. This may have performance implications.

[lovelyallen]

I've encountered this warning when programming a resnet.
my torch-directml version is 0.2.5.dev20240914

this is the code:

import torch
import torch_directml
from torch import *
from torch import nn
import polars
import numpy as np
from PIL import Image
from torch import optim
from torch.optim.nadam import NAdam
from torch.optim.adam import Adam
dml=torch_directml.device()
torch.set_default_device(dml)
X=polars.read_csv("tr_x.csv").to_numpy()
X=np.delete(X,[0],1)
X=np.reshape(X,[-1,30,1,28,28])
print(X.shape)
X=tensor(X,dtype=uint8)
Y=polars.read_csv("tr_y.csv").to_numpy()
Y=np.delete(Y,[0],1)
Y=np.reshape(Y,[-1,30])
Y=tensor(Y,device="cpu")
Y=nn.functional.one_hot(Y,26).to(dml)
class Block(nn.Module):
def init(self):
super(Block,self).init();
self.conv1=nn.Conv2d(32,32,3,padding=1)
self.conv2=nn.Conv2d(32,32,3,padding=1)
self.relu=nn.ReLU()
def forward(self,x):
r=clone(x)
x=self.conv1(x)
x=self.conv2(x)
x=self.relu(x)
x=r+x
return x
mod=nn.Sequential(
nn.Conv2d(1,32,5),
Block(),
Block(),
Block(),
Block(),
Block(),
Block(),
Block(),
Block(),
Block(),
Block(),
Block(),
Block(),
Block(),
Block(),
Block(),
Block(),
Block(),
Block(),
Block(),
nn.Flatten(),
nn.Linear(322424,128),
nn.ReLU(),
nn.Linear(128,26),
nn.Sigmoid()
).to(dml)
epoches=100
opt=NAdam(mod.parameters())
los=nn.CrossEntropyLoss().to(dml)
for _ in range(0,epoches):
total_loss=0
total_acc=0
for i in range(0,X.shape[0]):
x=X[i.item().int()].type(float32)
y=Y[i.item().int()].type(float32)
#print(y.shape)
pred=mod(x)
opt.zero_grad()
loss:Tensor=los(pred,y)
loss.backward()
opt.step()
if (i%100==0):
print(i)

any solutions?

(I am new to here and i dont know what to do with coping. sorry for that)

[AbdullahTrees]

The developers basically have to implement this function in the back-end, really hope they fix it soon. Adam optimizer is a very basic and fundamental part of AI/ML workloads and is really important for students like us.