*On single-machine and 4-machine 8-card V100 machines, model training is performed based on [PP-YOLOE-s](../../configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml). The model training time is as follows.
*We conducted model training on 3x8 V100 GPUs. Accuracy, training time, and multi machine acceleration ratio of different models are shown below.
Machine | mAP | Time cost
-|-|-
single machine | 42.7% | 39h
4 machines | 42.1% | 13h
| Model | Dataset | Configuration | 8 GPU training time / Accuracy | 3x8 GPU training time / Accuracy | Acceleration ratio |
* When the number of GPU cards for training is too large, the accuracy will be slightly lost (about 1%). At this time, you can try to warmup the training process or increase some training epochs to reduce the lost.
* The configuration files here are provided based on COCO datasets. If you need to train on other datasets, you need to modify the dataset path.
* For the multi-machine training process of `PP-YOLOE` series, the batch size of single card is set as 8 and learning rate is same as that of single machine.
