Open athul-bos-semi opened 6 days ago
@athul-bos-semi The example seems to use deprecated tt_lib
library. Which version of tt-metal are you using? It would be better to port it to the latest ttnn api.
I ported to v0.51.0, but ttnn.reshard does not work. The I ported it to main branch where reshard works, but convolution throws L1 memory error for code that was previously working.
but convolution throws L1 memory error for code that was previously working.
Can you please provide repro details on this? Since you say this regression is with the latest main
version, we need to look into that issue first.
You can download both the files from here and run the test_unet.py file to reproduce the errors.
I mean how to repro the conv L1 memory error that you get with the latest main
version of the repo, not the old version.
The main has already changed, now the error is different. Can you try and run these two files? I am now running on specifc versions to avoid confusion instead of running on main.
Describe the bug The program was working, until after Bilinear Upsampling stage. The very next convolution gets stuck in the _HWCommandQueue_writebuffer stage according to Tracy. I’m also attaching the results of Tracy for reference.
To Reproduce Steps to reproduce the behavior:
Expected behavior The program should run to completion.
Screenshots
Environment Information:
Additional context Use Tracy when checking. Also use L1 Buffer analyzer if available.