-
IREE currently lacks support for fusing `LinalgExt` operations with either other `LinalgExt` operations or standard `Linalg` operations..The current fusion implementation relies on indexing maps to de…
-
`+=`, `-=`, `*=` in https://github.com/tensorflow/swift-apis/blob/master/Sources/TensorFlow/Operators/Math.swift are not differentiable, but it should be pretty straightforward to make them differenti…
-
Currently the `Conv2d` layer in ChAI assumes that there must be a learnable bias for the output. (`bias`)
Pytorch support the option to turn this off. See https://pytorch.org/docs/stable/generated/t…
-
**Describe the bug**
Some files are missing the headers that they rely on, which means they cannot be included by themselves. This is "hidden" in most of the examples because they import many things a…
-
### 🚀 The feature, motivation and pitch
Can we add a `dtype` parameter to `count_nonzero()` to specify the output dtype?
Currently `count_nonzero()` outputs a int64 tensor. When working with large…
-
For example:
- when one of the operations in the reduction is not a tensor element, but the operation is commutative so it can be moved to the end.
-
Conv2d will shard the input tensor on its own, provided a sharding specification. By default it will attempt to use `HEIGHT_SHARDED`, but for some convolutions `BLOCK_SHARDED` or `WIDTH_SHARDED` might…
-
With the discussion in #1272 #1273 , I have been reminded of how uncomfortable I feel about controlled versions of bloqs with RIGHT or LEFT (allocating / de-allocating) registers.
There are parts o…
-
When doing the evaluation, the following error accurs:
```bash
tensorflow.python.framework.errors_impl.InvalidArgumentError: 2 root error(s) found. …
-
**Describe the bug**
Following the instructions in [`examples/mistral`](https://github.com/microsoft/Olive/tree/main/examples/mistral) does not result in a quantized onnx model. After running the wor…