jit.rst

TorchScript Support

TorchScript is a way to create serializable and optimizable models from :pytorch:`PyTorch` code. Any TorchScript program can be saved from a :python:`Python` process and loaded in a process where there is no :python:`Python` dependency. If you are unfamilar with TorchScript, we recommend to read the official "Introduction to TorchScript" tutorial first.

Converting GNN Models

Note

From :pyg:`PyG` 2.5 (and onwards), GNN layers are now fully compatible with :meth:`torch.jit.script` without any modification needed. If you are on an earlier version of :pyg:`PyG`, consider to convert your GNN layers into "jittable" instances first by calling :meth:`~torch_geometric.nn.conv.MessagePassing.jittable`.

Converting your :pyg:`PyG` model to a TorchScript program is straightforward and requires only a few code changes. Let's consider the following model:

import torch
import torch.nn.functional as F
from torch_geometric.nn import GCNConv

class GNN(torch.nn.Module):
    def __init__(self, in_channels, out_channels):
        super().__init__()
        self.conv1 = GCNConv(in_channels, 64)
        self.conv2 = GCNConv(64, out_channels)

    def forward(self, x, edge_index):
        x = self.conv1(x, edge_index)
        x = F.relu(x)
        x = self.conv2(x, edge_index)
        return F.log_softmax(x, dim=1)

model = GNN(dataset.num_features, dataset.num_classes)

The instantiated model can now be directly passed into :meth:`torch.jit.script`:

model = torch.jit.script(model)

That is all you need to know on how to convert your :pyg:`PyG` models to TorchScript programs. You can have a further look at our JIT examples that show-case how to obtain TorchScript programs for node and graph classification models.

Creating Jittable GNN Operators

All :pyg:`PyG` :class:`~torch_geometric.nn.conv.MessagePassing` operators are tested to be convertible to a TorchScript program. However, if you want your own GNN module to be compatible with :meth:`torch.jit.script`, you need to account for the following two things:

1. As one would expect, your :meth:`forward` code may need to be adjusted so that it passes the TorchScript compiler requirements, e.g., by adding type notations.

2. You need to tell the :class:`~torch_geometric.nn.conv.MessagePassing` module the types that you pass to its :meth:`~torch_geometric.nn.conv.MessagePassing.propagate` function. This can be achieved in two different ways:

   1. Declaring the type of propagation arguments in a dictionary called :obj:`propagate_type`:

   from typing import Optional
   from torch import Tensor
   from torch_geometric.nn import MessagePassing
   
   class MyConv(MessagePassing):
       propagate_type = {'x': Tensor, 'edge_weight': Optional[Tensor] }
   
       def forward(
           self,
           x: Tensor,
           edge_index: Tensor,
           edge_weight: Optional[Tensor] = None,
       ) -> Tensor:
           return self.propagate(edge_index, x=x, edge_weight=edge_weight)

   2. Declaring the type of propagation arguments as a comment inside your module:

   from typing import Optional
   from torch import Tensor
   from torch_geometric.nn import MessagePassing
   
   class MyConv(MessagePassing):
       def forward(
           self,
           x: Tensor,
           edge_index: Tensor,
           edge_weight: Optional[Tensor] = None,
       ) -> Tensor:
           # propagate_type: (x: Tensor, edge_weight: Optional[Tensor])
           return self.propagate(edge_index, x=x, edge_weight=edge_weight)

   If none of these options are given, the :class:`~torch_geometric.nn.conv.MessagePassing` module will infer the arguments of :meth:`~torch_geometric.nn.conv.MessagePassing.propagate` to be of type :class:`torch.Tensor` (mimicing the default type that TorchScript is inferring for non-annotated arguments).