-
FP16 and use_amp support
-
[Docs](https://azure.github.io/MS-AMP/docs/introduction)
MS-AMP would allow us to also store the weights in FP8, allowing for larger models to be trained on smaller hardware, as right now the weigh…
-
I'm adding 2 additional SONOS AMPs to my network this summer. The first for a newly constructed fire pit and the second for the recently buried trampoline (rock speakers). In planning out the wiring r…
-
Failure link
------------
From [an internal build](https://hyc-runtimes-jenkins.swg-devops.com/job/Test_openjdk23_j9_extended.openjdk_ppc64le_linux_testList_0/7/)(`sles12le-rtp-rt5-1`):
```
open…
-
### 🐛 Describe the bug
torchbench_amp_bf16_training
xpu train tacotron2
W0803 05:40:15.849000 2621977 torch/_inductor/utils.py:1357] [5/0_1] DeviceCopy in input program
W08…
-
### 🐛 Describe the bug
Model list:
- [ ] `resnest101e`
E0804 04:47:13.347000 901510 torch/_dynamo/utils.py:1558] RMSE (res-fp64): nan, (ref-fp64): 0.00000 and shape=torch.Size([128]). res.dtype: …
-
Currently, Explore Metrics only supports the Grafana core Prometheus data source. Because we created the Prometheus libraries and were able to create other vendor Prometheus, this issue is to support …
-
Thanks for sharing the well organized code. May I ask how to use SAM under torch AMP (mixed precision training)?
-
The RTX 3080 Ti is set with half_precision_backend: 'auto' and fp16=True, but it’s not effective and memory usage is not reduced. What could be the possible reasons?
-
Use [AMPHP](https://github.com/amphp/amp?tab=readme-ov-file) to go async