Kai Londenberg comments

Results 27 comments of


                                            Kai Londenberg

[BUG] Misaligned address when running GEMM with SM90 EVT-based epilogue

Just as additional info: Adding -DNDEBUG and -O3 and removing -g and -lineinfo from the build flags does not make a difference here in my tests.

[BUG] Misaligned address when running GEMM with SM90 EVT-based epilogue

If remember correctly, A and B are both square matrices of Same shape here. You can find the details in the gist, where you will find a standalone source code...

[BUG] Misaligned address when running GEMM with SM90 EVT-based epilogue

No, this is still a bug as far as I can tell. It's not urgent, though, since we're not using auxiliary inputs anymore.

Does Concatenate order matters?

thanks for your report, we will try to reproduce the issue and then get back to you.

error during inferencing: Error: Constant embeddings_token_embedding_weight was not set! Set the value with set_constant.

Thanks for your report, we will look into it.

[Inductor max autotune] Make autotune_select_algorithm more robust

@int3 I noticed you wrote a diff that's doing something similar. Will take a look and see that this one is still compatible with your changes.

[Inductor max autotune] Make autotune_select_algorithm more robust

@pytorchbot merge

[Inductor max autotune] Make autotune_select_algorithm more robust

@pytorchbot merge

[Inductor cutlass backend] Fix cutlass_utils.get_max_alignment() for strided layouts.

@pytorchbot merge

[Inductor cutlass backend] Fix cutlass_utils.get_max_alignment() for strided layouts.

@pytorchbot merge