Daniel Garvey
Daniel Garvey
@GMNGeoffrey can you take a look? Is there a way to test this besides merging it and running it manually?
@mariecwhite Something I mentioned in our meeting in the other day was incorrect. What I added here does not run for 3 hours, as it doesn't generate new artifacts, just...
> > @mariecwhite Something I mentioned in our meeting in the other day was incorrect. What I added here does not run for 3 hours, as it doesn't generate new...
@mariecwhite what set would you like run?
Yeah, we're going to relax tolerance when TF32 is enabled. Any changes would be on the NVIDIA side, so nothing required from the IREE side afaik. Thanks for the support!
Can you squash the commits
somehow reopening this added a prior commit, I'll fix this tomorrow
In general larger image sizes are going to take longer/use more resources, so this is largely expected. A fix went in recently that should allow for switching from img2img back...
it started its life as `view(1, 1, key_size, key_size)`, I actually circumvented the error by rewriting it with unsqueezes. I suppose the problem is with double dynamic dims like that...
I am actually coming via fx_importer, but I dont think the view lowering takes the strict_symbolic_shapes into account? here's a link: https://huggingface.co/Qwen/Qwen-7B-Chat/blob/main/modeling_qwen.py#L505 the other offending views are the same as...