项目文件夹

文件

T

文件历史

Chang Liu fce9614089 [Example][Refactor] Refactor graphsage multigpu and full-graph example (#4430 )

* Add refactors for multi-gpu and full-graph example

* Fix format

* Update

* Update

* Update

2022-09-05 09:05:56 +08:00

multi_gpu_graph_prediction.py

2022-08-19 09:38:19 +08:00

multi_gpu_node_classification.py

2022-09-05 09:05:56 +08:00

README.md

2022-08-19 09:38:19 +08:00

Multiple GPU Training

pip install torchmetrics

Run with following (available dataset: "ogbg-molhiv", "ogbg-molpcba")

python3 multi_gpu_graph_prediction.py --dataset ogbg-molhiv

* ogbg-molhiv: ~0.7965
* ogbg-molpcba: ~0.2239

We test scalability of the code with dataset "ogbg-molhiv" in a machine of type Amazon EC2 g4dn.metal , which has 8 Nvidia T4 Tensor Core GPUs.

GPU number	Speed Up	Batch size	Test accuracy	Average epoch Time
1	x	32	0.7765	45.0s
2	3.7x	64	0.7761	12.1s
4	5.9x	128	0.7854	7.6s
8	9.5x	256	0.7751	4.7s

Run with following on dataset "ogbn-products"

python3 multi_gpu_node_classification.py

Test Accuracy: ~0.7632