bhack
bhack
As I mentioned the repro is at https://github.com/pytorch/pytorch/issues/121504 What is a full error trace to debug this?
On the H100 with pytorch nightly ```python Traceback (most recent call last): File "/workspace/tools/eval.py", line 135, in main() File "/workspace/tools/eval.py", line 130, in main main_worker(0, cfg, enable_amp=args.amp) File "/workspace/tools/eval.py", line...
@ezyang For the problem instead on the [A100/L4](https://github.com/pytorch/pytorch/issues/126614#issuecomment-2118873233) do you know where `index_put` is going to require `nonzero` op? Is it derived from: https://github.com/pytorch/pytorch/blob/0756f9f5fd0c58e2205da41da39dd5374f81404f/aten/src/ATen/native/TensorAdvancedIndexing.cpp#L8-L9
Ok so in this case we are going to hit again https://github.com/NVIDIA/cccl/issues/1422 for some specific inputs. But do you know instead what it is [happening on H100](https://github.com/pytorch/pytorch/issues/126614#issue-2303987817) ?
Tested again on `20240530` nightly the error https://github.com/pytorch/pytorch/issues/126614#issuecomment-2122567229 on H100 is still there.
I don't think it will merged. See https://github.com/facebookresearch/co-tracker/issues/31
@nikitakaraevv It would be nice if with the new version we could enable/control a strategy to automatically allocate new points/grids on new incoming contents on the fly. As new incoming...
Any news on this?
Thanks, was the bug also in the original version?
Do you think that it has also an impact for https://github.com/facebookresearch/co-tracker/issues/15 ?