-
For a CTMRG algorithm I need `linalg.Qr` with a linear operator as an input instead of a `Tensor` or a `UniTensor`. This is crucial for the runtime.
Currently, only `Arnoldi` and `Lanczos` (in diffe…
-
### 🐛 Describe the bug
I'm using the script below to export the Flux Transformer model to ONNX using torch.onnx.dynamo_export(). However, I run into a TypeError relating to an attribute type.
The…
-
I am seeing a strange bug where the datatype of a reshape and transpose ops is affecting the error of a much further downstream op.
Cast output of linear0 from bfloat16->float before reshape/transp…
-
The torchvision Gaussian function runs on the CPU even if your tensor is on the GPU. Here's a blur function that guarantees the blur will run very quickly on the GPU. Some adaptation for the tensor sh…
-
Hello,
I've been experimenting with the pre-trained model you've uploaded, and I've noticed an issue. Regardless of the input, the model consistently predicts every image to be label 4. This behavi…
-
The `*` op is supposed to concatenate a shape tensor to itself. However, if we use the resulting shape tensor with a fill ops, we get a failure:
```py
>>> a = tp.Tensor([1, 2, 3])
>>> tp.ones(a.shape[…
-
Hi there,
I am reading through some of the documentation and the section `FEValues` was not so straightforward to read for somebody coming from a background where tensors are not common. I think m…
-
Now that we have localizations of rings #2089 we should introduce localizations of R-modules. This will require two things:
1. [ ] Tensor products of modules #1926.
2. [ ] Showing that `rng_locali…
-
Some ops can produce scalar values, like Matmul vector-vector product:
https://github.com/tenstorrent/tt-mlir/pull/745#discussion_r1764954874
Today, we are converting scalar values into 1D tensors…
-
**Target:** Figure out if by default Torch `scaled_dot_product_attention` when executing in Hopper architecture, thus `PLATFORM_SUPPORTS_CUDNN_ATTENTION` and `SM90OrLater` evaluate to `True`, executes…