-
Using an Triton 2.0.0, Pytorch 2.0.0, Python 3.9.16, Cuda 11.6 on a pc running Centos release 7.4.1708 with an nvidia A100. I using the `matmul` and `blocksparse/matmul` ops in [https://github.com/o…
-
### 🚀 The feature, motivation and pitch
I need to observe the strength of attention for each element in the sequence with all the elements in the same sequence. This is called the attention map in …
-
Let me post a test @mschreter and I wrote last year (Nov 9, 2022) and which reminds me of the problematic configuration we have here. Left cell is refined and linear and right cell is not refined and …
-
I've been going back and forth with RoBERTa and Longformer for classification. My typical use case is quite sporadic, as most of my documents are around 300 tokens, but occassionally I get massive 10…
-
### 📚 The doc issue
requesting to add `FlopCounterMode` documentation
and to include the following notes from these discussion:
1. https://pytorch.slack.com/archives/C3PDTEV8E/p17127855391758…
-
Hi all,
Would it make sense and would it be possible to add options into the blocksqp interface for providing custom callbacks for the computation of the lagrangian of the hessian, gradient of the…
tsijs updated
5 years ago
-
**System information**
- TensorFlow version (you are using): 2.5.0
- Are you willing to contribute it (Yes/No): Yes
**Motivation**
Deciding on where to have high filter/channel counts in con…
-
It's a minor issue, but I accidentally evaluated a large sparse matrix (~500M rows) in REPL and it caused a ~5-second freeze. Not sure what's happening under the hood but I assume it tries to output a…
-
Not sure if this is intended behavior, but it looks like there might be an issue with concatenation based on the following test.
Code:
```
class TestModule(nn.Module):
def __init__(self, in_…
-
### Description of the enhancement or error report
Preallocation of matrix memory is crucial for good performance for large problems. There are more details at https://www.mcs.anl.gov/petsc/documenta…