What size models do you build in PyTorch?
Are you experimenting with small networks or building large, multi-block architectures? This helps optimize UI performance and features.
Call for PyTorch-heavy testers
If your day-to-day work is deep in PyTorch and you’re willing to break early builds, I want to hear from you. Leave a comment and I’ll reach out.
Roadmap preview. What priorities would you change?
Planned: • Drag-and-drop layer editing • Reusable blocks (ResNet-style) • Automatic shape tracking • PyTorch nn.Module export Which of these should be higher/lower priority? What’s missing?
What makes PyTorch model wiring painful?
Even experienced users hit snags with shape mismatches and repetitive boilerplate. Where does it break down for you?
Show me your current PyTorch workflow
Do you sketch models, copy older projects, or code directly in VS Code? Understanding your workflow helps me design better drag-and-drop components.
Feature request thread (PyTorch first)
NetSmith is PyTorch-first. If you could add one feature tomorrow for PyTorch models, what would it be?
What PyTorch export format do you want?
NetSmith currently exports clean PyTorch modules. I can expand the export options — what would you prefer? Standard nn.Module Lightning-style modules Scriptable modules Custom template support
What’s the biggest pain point in building PyTorch models?
When you’re wiring up a new architecture in PyTorch, what slows you down the most? Layer selection? Shape debugging? Boilerplate? Something else?