cora-codes

Tensor Methods are Broken ("RuntimeError: Cannot call @triton.jit'd outside of the scope of a kernel")

3

### Describe the bug When using method calls on tensors (e.g., `tensor.sum(axis=0)` instead of `tl.sum(tensor, axis=0)`), I get the following runtime exception: `RuntimeError: Cannot call @triton.jit'd outside of the scope...

Error when building PyTorch from source

3

### Question I'm using the latest version of UV. `uv add --editable external/pytorch/` Gives the following error: ``` × Failed to build `torch @ file:///home/ubuntu/pytorch` ├─▶ The build backend returned...

As the title states, we've noticed kernels using dynamic parallelism do not show up in the profiler. It would be a nice quality of life change if they did.