Ptf ReLU

ReLU is the boring, correct default of deep learning - max(0, x), nothing more - and Ptf ReLU packages it as a callable component for the pack's no-code model building. It has no inputs; what you get out is the activation function itself, ready to be dropped into a Pt Apply Function node or a model closure. If you're assembling a network from parts and you want the activation between layers, this is the one you'll grab nine times out of ten.

How it works

The node returns torch.nn.functional.relu - the raw function, not a module. Output type is PTCALLABLE, and you apply it to an actual tensor with Pt Apply Function (tensor + callable in, transformed tensor out) or as the closure in model-building nodes that accept one. Because it's a plain function, it composes cleanly: chain ReLU after a linear layer, before the next one, exactly like writing F.relu between nn.Linear layers in raw PyTorch.

ReLU's appeal is that it's cheap, it works, and its gradient vanishes much less than sigmoid/tanh for positive inputs. Its one real weakness - dead neurons on the zero side for negative inputs - is the reason packs like this also offer Ptf SiLU and Ptf Softplus, which are smooth variants. But for a first no-code network, ReLU is the right call.

The one output

• PTCALLABLE - the ReLU function. No inputs at all.

The only decision you're making is which activation node to drop in. If you're not sure: start with ReLU for hidden layers, keep Ptf Softmax or Ptf Log Softmax for the output layer when you're classifying.

Where people get burned

• Forgetting that it's a callable - you can't feed a tensor into this node, because it has no tensor input. Wire the PTCALLABLE into Pt Apply Function or a closure, then feed the tensor there.
• Expecting an nn.Module - this is the functional form. Some nodes want a module; read the socket, and use the pack's layer nodes (e.g. Add Linear Layer-style) when a module is required.

Installing it

It's in ComfyUI-Pt-Wrapper:

• ComfyUI Manager → search "ComfyUI-Pt-Wrapper" → Install → restart.
• Or cd ComfyUI/custom_nodes && git clone https://github.com/HowToSD/ComfyUI-Pt-Wrapper and restart.

The pack installs a heavy ML stack (transformers, datasets, peft, accelerate, scikit-learn, scipy, gensim, sentencepiece), so first boot is slow. No model downloads needed. It's a niche educational pack by HowToSD with little community presence - the repo's docs/reference/ is the authoritative doc.