Cellulose provides some support for uploading and visualizing your PyTorch models via our Python SDK.

By using our APIs, providing an input tensor (shapes) and a torch.nn.Module, we’ll export it to an ONNX model and upload it to the cloud automatically. We’ll walk through setup instructions and a basic example in the next section.

While this still has an ONNX intermediary step, we plan to also add support for several other PyTorch IRs like TorchScript (automatically uploading these models as .pt or .ts), or more recent FX IRs which are heavily used by ML engineers to trace and modify op nodes without touching too much model code. Stay tuned for these releases!

Have questions / need help?

Please reach out to support@cellulose.ai, and we’ll get back to you as soon as possible.