-
Dear authors,
First of all, thanks a lot for this very great work which does inspire us.
I have a question w.r.t the [calculation of the pointer_logits](https://github.com/manycore-research/PlankAs…
-
For some application, I need to get gradients for each elements of a sum.
I am aware that this issue has already been raised previously, in various forms ([here](https://discuss.pytorch.org/t/gradien…
-
## 🚀 Feature
This RFC proposes to extend Pytorch Autocast to support datatype of BFloat16 on CPU device. Autocast provides easy-to-use methods for mixed precision which can use the low precision data…
-
Remove all custom matmuls/bmms and replace with ttnn matmuls/linears. There might be minor host perf degradation since some of the asserts are in python now.
List:
- [x] bert matmuls/bmms
- [ ] …
-
I wanted to change 'imsize' that is paramater.
So, I changed 'imsize' from 64 to 128 but I got the following error message.
`AttributeError: 'Discriminator' object has no attribute 'l4'`
Could yo…
ghost updated
5 years ago
-
For some of the models we are developing in https://github.com/venpopov/bmm/, we have run into an issue that the models need us to specify initial values, otherwise sampling cannot initiate. I know ho…
-
Canonical ops: https://docs-preview.pytorch.org/90644/ir.html
The list below is an incomplete list of canonical ops + other ops ordered by usage. canonical ops are marked with `canonical`. Be sure…
-
-
The code causing segmentation fault is `torch.bmm` op. So does ncnn support `torch.bmm` or `torch.mm` which is `GEMM` or `MatMul` in ONNX?
Here is the target onnx file:
[mm.zip](https://github.co…
-