Following pattern in code seems to break dynamo graph tracing

import torch
import torchdynamo

from torch import nn

class TestModel(nn.Module):
    def __init__(self) -> None:
        super().__init__()
        self.linears = nn.Sequential(
            nn.Linear(100, 100),
            nn.Linear(100, 100),
            nn.Linear(100, 100),
            nn.Linear(100, 100),
        )

    def forward(self, x):
        all_but_last, last = self.linears[:-1], self.linears[-1]
        x = all_but_last(x)
        x = last(x)
        return x


# @torchdynamo.optimize('eager')
def toy_example(model, input):
    return model(input)

model = TestModel().cuda()
x = torch.rand((8, 100), device='cuda')

for _ in range(10):
    toy_example(model, x)

Its okay if the answer is, this is not supported but just wanted to raise it as fyi

[voznesenskym]

@voznesenskym we’re still using the dynamo issue tracker

My mistake. Apologies @ngoyal2707

[voznesenskym]

gotcha 😃 thanks - can you port this issue to pytorch? Just realized its in dynamo, and this repo is effectively closed.