cccclai
cccclai
what does the graph look like after you run `to_backend`?
what does you code look like around call-site `to_backend`?
Link to landed trunk PR: * https://github.com/pytorch/executorch/pull/3490 Link to release branch PR: * https://github.com/pytorch/executorch/pull/3506 Criteria category and reasons: A bug for the backend delegate, hopefully it can be cherry picked...
Did you set the environment correctly? > $QNN_SDK_ROOT refers to the root of Qualcomm AI Engine Direct SDK, i.e., the directory containing QNN_README.txt. > $ANDROID_NDK refers to the root of...
From `PrecomputeFreqsCIS`, it looks like `seq_len` is real user input, what would you like to do with the constant data? Since it's user input, I expect it to be also...
Hmm I think depending on your code implementation, it can either be tensor or int...I'm not quite sure the goal for your backend. Are you trying to lower some operators...
I think it's possible that the partitioner is a bit off, and the graph after `to_backend` is a bit off. For this example, the graph is ``` class GraphModule(torch.nn.Module): def...
> What do you mean the partitioner or the graph is off? It means being invalid? Yeah I'd guess so. > Yes. Then, should I set y as a buffer...
Seems like that's the link to the stable branch. Did you check out the stable branch code or main branch? The main branch doc is https://pytorch.org/executorch/main/running-a-model-cpp-tutorial.html
I think we can use python virtual environment, but will let @larryliu0820 and @dbort confirm