[Feature] Support CPU training #7016
Conversation
Codecov Report
@@ Coverage Diff @@
## dev #7016 +/- ##
==========================================
+ Coverage 62.35% 62.39% +0.03%
==========================================
Files 327 329 +2
Lines 26129 26176 +47
Branches 4424 4432 +8
==========================================
+ Hits 16293 16332 +39
- Misses 8969 8975 +6
- Partials 867 869 +2
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
|
The bug about non-distributed multi-GPU training is removed to #7019 |
| **Note**: | ||
|
|
||
| We do not recommend users to use CPU for training because it is too slow. We support this feature to allow users to debug on machines without GPU for convenience. |
There was a problem hiding this comment.
Should have used the "note" wrapper
```{note}
```
|
[question] If I want to use mmdet/apis/train.py , how to understand "device_ids=cfg.gpu_ids"? |
* Modify docs * Support CPU training * Modify docs * Modify Chinese docs
* Modify docs * Support CPU training * Modify docs * Modify Chinese docs
* Modify docs * Support CPU training * Modify docs * Modify Chinese docs
We support CPU training.
The model is default put on cuda device, we modify this setting so that if there are no cuda devices, the model will be put on cpu.
This PR is based on open-mmlab/mmcv#1621, we should merge this PR after 1621.
Now we can use the CPU to train/debug our model and test our model with batch size >=2. Before running the program we need to export CUDA_VISIBLE_DEVICES=-1 to disable GPU visibility.