ROCm / ROCm

AMD ROCm™ Software - GitHub Home

Home Page:https://rocm.docs.amd.com

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

[Issue]:

hbfreed opened this issue · comments

Problem Description

After installing 6.1, when I use torch.compile, I get the following traceback (I can put the whole thing in a pastebin, it was just long):
If I don't compile, the train script runs just fine. Happy to provide any more details to get things sorted out.

File "/usr/lib/python3.10/concurrent/futures/_base.py", line 403, in __get_result
raise self._exception
torch._dynamo.exc.BackendCompilerFailed: backend='inductor' raised:
RuntimeError: Internal Triton PTX codegen error:
ptxas fatal : Value 'sm_110' is not defined for option 'gpu-name'

Operating System

22.04.4 LTS (Jammy Jellyfish)

CPU

EPYC 7402P

GPU

AMD Radeon RX 7900 XTX

ROCm Version

ROCm 6.0.0

ROCm Component

No response

Steps to Reproduce

No response

(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support

No response

Additional Information

No response

Update: just downgraded to 6.0.2, it did not fix this problem. I've tried both regular triton and the triton nightly, for context there.

I fixed this, I just fully reinstalled torch and triton, making sure to uninstall the triton that torch packages with it.