Studio Perforate your model.
Perforated is the efficiency layer for machine learning that adds artificial dendrites to your PyTorch layers. Perforated Studio is the coding agent and dashboard that helps anyone perforate their models to achieve higher accuracy with greater efficiency.
$ curl -fsSL .../bootstrap.sh | sh
- Perforating model
✓ wrapped Linear(512, 512) -> 3 dendrites ready
HOW IT WORKS
Three steps. No ML engineering experience required. No new dashboard to remember.
01
Install
One docker instance is installed to manage all of your perforated projects. The skills are automatically plugged into your coding agent.
02
Say "Perforate my model"
Claude wraps your existing PyTorch Modules and configures them to automatically add dendrites. No need to rewrite your code.
03
Train, and watch it grow
The Dashboard opens in your browser and streams the run live, including score, switches, and every dendrite as it's added.
NOT A BLACK BOX
It's your model the whole way through.
Runs on your machine
One docker container per dev machine. Nothing about your model or your data leaves it.
Native PyTorch
Perforated modules wrap the existing PyTorch layers you already use. It remains a
normal nn.Module
Claude code skills, not a tab to remember
Perforate, train, and analyze live from where you work.
Open architecture
Inspect perforated modules directly, any time: the weights are just weights
THE DASHBOARD
This is what perforating a neural network looks like.
Claude opens it for you when training starts. It's the visual representation, not the product. The model still trains from your script.
Define objective
Setup compute
Training — epoch 14, switch 2
Training complete
Score
val
train
Dendrites
epoch 12 - val 0.847
epoch 13 - val 0.851
switch - dendrite added