-
There seem to be inconsistencies in TTNN IR op layout inputs and outputs. Currently all op inputs are set to TILE layout, while all outputs are ROW_MAJOR. Example:
```mlir
#device = #tt.device, l1Ma…
-
# Purpose/Motivation
Provides a means to upload or download tensor data to/from a created `MLBuffer` via #542 . This is a sub-issue of https://github.com/webmachinelearning/webnn/issues/482.
# …
-
### 📚 The doc issue
[The doc](https://pytorch.org/docs/stable/generated/torch.linalg.norm.html) of `linalg.norm()` says there is `A` parameter for `linalg.norm()` as shown below:
> torch.linalg.no…
-
RuntimeError: The expanded size of the tensor (1024) must match the existing size (1280) at non-singleton dimension 3. Target sizes: [1, 3, 512, 1024]. Tensor sizes: [3, 512, 1280]
![image](https…
-
### Issue Type
Feature Request
### OS
Linux
### onnx2tf version number
1.20.0
### onnx version number
1.16.2
### onnxruntime version number
1.19.2
### onnxsim (onnx_simplifier) version numbe…
-
I am trying to benchmark NKI code generated from the corresponding PyTorch code. My workflow is as follows:
1. I run the PyTorch kernel and extract the `.pb` file
2. I compile the NKI kernel using:
…
-
输入
c
name: c
tensor: float32[c_dynamic_axes_1,c_dynamic_axes_2,768]
f0
name: f0
tensor: float32[1,f0_dynamic_axes_1]
mel2ph
name: mel2ph
tensor: int64[1,mel2ph_dynamic_axes_1]
uv
name: uv…
-
### 🐛 Describe the bug
### 🚀 The feature, motivation and pitch
This is moving "issue 2" from https://github.com/pytorch/pytorch/issues/130015 to be tracked separately.
**Context**:
While using…
vkuzo updated
3 weeks ago
-
I use tons of torch.Tensor.to methods in my model for moving tensor between cpu/cuda, but the method is not supported in torch2trt currently. Will it ever be supported in the future release?
torch.Te…
-
**Describe the bug**
I discovered the bug while testing because I wanted to try using the `dataflow` CB instead of `intermed`.
There are no issues when using bfloat16 tensors.
However, when using…