-
## Logger
![image](https://user-images.githubusercontent.com/10796970/57189045-dc393100-6f3b-11e9-82ff-29641523e0df.png)
![image](https://user-images.githubusercontent.com/10796970/57189342-405df4…
-
https://twitter.com/tempus42/status/1034924269380284418
-
torch native amp + apex amp
-
firstly installed torch 1.9
downgraded to 1.8 after seeing solutions on stack
Current :-
torch==1.8.1+cu111
torchaudio==0.8.1
torchmetrics==1.0.1
torchvision==0.9.1+cu111
```
Traceback (m…
-
Hi, I started a group of processes to perform allreduce operations. Each process started another thread to call `ncclCommAbort` at certain timepoint.
It is expected that all processes will eventua…
-
PerFlow is a great tool and we are interested in code implementation. However, we encountered some missing files while trying to reproduce it, such as: pag.gml, mpi_mpag.gml, output.json, pag_to_mpag.…
-
**The current situation**
- It unfortunately looks like @pkittenis is incommunicado since late 2022.
- I am an GitHub org member, but don't have admin access
- Therefore nothing can be merged i…
-
Start a repo and fork it.
-
## 🚀 Feature
I propose adding instead of batch size a dictionary with batch size per GPU, for example {"cuda0":4, "cuda1", 6}
### Motivation
I have gv100 (32gb) and 3090 (24gb). using the cur…
-
## 🐛 Bug
In commit `6e14209185c2b2100f3e515ee6782597673bb921` on pytorch_lightning from Feb 17, the `use_ddp` property was removed from AcceleratorConnector.
In commit `b29b07e9788311326bca4779d…