Multigpu cleanup for readme and name (#8577)

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Akihiro Nitta <[email protected]>
pyg-team · Dec 8, 2023 · 6f5fabe · 6f5fabe
1 parent 90e7604
commit 6f5fabe
Show file tree

Hide file tree

Showing 4 changed files with 21 additions and 0 deletions.
diff --git a/examples/multi_gpu/README.md b/examples/multi_gpu/README.md
@@ -0,0 +1,21 @@
+# Examples for Distributed Training
+
+## Examples with Nvidia GPUs
+
+| Example                                                                          | Scalability | Description                                                                                                                                                                         |
+| -------------------------------------------------------------------------------- | ----------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| [distributed_batching.py](./distributed_batching.py)                             | single-node | Example for training GNNs on multiple graphs.                                                                                                                                       |
+| [distributed_sampling.py](./distributed_sampling.py)                             | single-node | Example for training GNNs on a homogeneous graph with neighbor sampling.                                                                                                            |
+| [distributed_sampling_multinode.py](./distributed_sampling_multinode.py)         | multi-node  | Example for training GNNs on a homogeneous graph with neighbor sampling on multiple nodes.                                                                                          |
+| [distributed_sampling_multinode.sbatch](./distributed_sampling_multinode.sbatch) | multi-node  | Example for submitting a training job to a Slurm cluster using `distributed_sampling_multi_node.py`.                                                                                |
+| [papers100m_gcn.py](./papers100m_gcn.py)                                         | single-node | Example for training GNNs on a homogeneuos graph.                                                                                                                                   |
+| [papers100m_gcn_multinode.py](./papers100m_gcn_multinode.py)                     | multi-node  | Example for training GNNs on a homogeneous graph on multiple nodes.                                                                                                                 |
+| [taobao.py](./taobao.py)                                                         | single-node | Example for training GNNs on a heterogeneous graph.                                                                                                                                 |
+| [model_parallel.py](./model_parallel.py)                                         | single-node | Example for model parallelism by manually placing layers on each GPU.                                                                                                               |
+| [data_parallel.py](./data_parallel.py)                                           | single-node | Example for training GNNs on multiple graphs. Note that `torch.nn.DataParallel` is slow and discouraged ([pytorch/pytorch#65936](https://github.com/pytorch/pytorch/issues/65936)). |
+
+## Examples with Intel GPUs (XPUs)
+
+| Example                                                      | Scalability            | Description                                                              |
+| ------------------------------------------------------------ | ---------------------- | ------------------------------------------------------------------------ |
+| [distributed_sampling_xpu.py](./distributed_sampling_xpu.py) | single-node, multi-gpu | Example for training GNNs on a homogeneous graph with neighbor sampling. |
diff --git a/...gpu/singlenode_multigpu_papers100m_gcn.py → examples/multi_gpu/papers100m_gcn.py b/...gpu/singlenode_multigpu_papers100m_gcn.py → examples/multi_gpu/papers100m_gcn.py
diff --git a/..._gpu/multinode_multigpu_papers100m_gcn.py → ...les/multi_gpu/papers100m_gcn_multinode.py b/..._gpu/multinode_multigpu_papers100m_gcn.py → ...les/multi_gpu/papers100m_gcn_multinode.py
diff --git a/examples/multi_gpu/taobao_multigpu_large.py → examples/multi_gpu/taobao.py b/examples/multi_gpu/taobao_multigpu_large.py → examples/multi_gpu/taobao.py