Stella Laurenzo
Stella Laurenzo
I think it is a good idea to pursue. There was really no reason it wasn't set up this way except for the early stage of the project. I'm happy...
Hey Sean, I obviously don't want to go against the community consensus here, but I do wonder if we are holding a bit of a different standard vs the rest...
How about to unblock progress, you all just create a `tcp` branch in the torch-mlir repo. Work there, and review each patch per usual guidelines as if it is going...
sg. Just keep the history clean and reviewed. Then assuming we decide to go forward with it at a future checkpoint, we accept it in full (applying any additional feedback/patched...
I can't commit to doing it regularly but am open to fill in if anyone needs to cancel or handoff. Maybe have a list of backups?
I think there are two issues here: 1. py_binary is enforcing a hard-coded list of file extensions when linkshared = True. "pyd" isn't on the list. I would advocate for...
I'm still seeing patches in OSS projects that work around this in gross ways so i think is still a problem. Please don't close unless if fixed.
Such a patch would be a a stopgap for this very specific case, but I can't think of any reason why the build tool should be enforcing anything here. If...
+1 - most downstreams are doing this (IREE, torch-mlir, Jax that I know of). Torch-mlir is probably the simplest at this point, if looking for a template.
The problem is that I don't think it can be dynamically 1. That is illegal at the pytorch level. (The assert is correct)