-
Notifications
You must be signed in to change notification settings - Fork 3.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[Bug] Missing PackedFunc with Specific Transformation Sequence in Relax Module #17357
Comments
This is a bit of a bug and a bit of an ordering dependency.
There's a couple of options for short-term fixes, and a couple of options for long-term fixes.
Unfortunately, I don't have time to implement the medium/long term solutions at the moment, but could help guide somebody in their implementation if there's interest. |
Thank you very much for your thorough analysis and explanation of the root cause of the bug, as well as the detailed guidance on how to address it. Unfortunately, I'm not too familiar with the relax source code, which means I might struggle with submitting a PR to fix this myself. I do hope someone with the right expertise and interest can pick this up. Thanks again for all your help, and I'm looking forward to seeing this issue tackled by the community! |
I encountered an issue while running a Relax module with a specific transformation sequence. Specifically, when
FuseTIR()
is applied once, the VM fails to find the PackedFuncfused_relax_nn_attention_cutlass_gv
. However, when theFuseTIR()
optimization is applied again beforeAllocateWorkspace()
, the problem disappears.Expected behavior
The script is expected to run successfully without errors.
Actual behavior
InternalError: Check failed: (func.defined()) is false: Error: Cannot find PackedFunc fused_relax_nn_attention_cutlass_gv in either Relax VM kernel library, or in TVM runtime PackedFunc registry, or in global Relax functions of the VM executable
Steps to reproduce
The following script reproduces the issue:
Any guidance on whether this is a bug or a known order dependency would be greatly appreciated.
@Lunderberg
The text was updated successfully, but these errors were encountered: