—Image input
[B, 3, 128, 128]no weights · 192 KB
Budgets & global pooling
Defect detection on the line
A production line photographs every part. The classifier must run on the line's own CPU between parts, so it gets a hard compute budget — and the plant may swap in a higher-resolution camera next year without retraining you a new head.
Classify a part as sound or defective inside a 500K-parameter and 50M-FLOP budget. Choose the collapse step that keeps the head independent of image size.
The meter
What this architecture costs
[B, 3, 128, 128]Parameters
0
FLOPs / example
0
Activations
192 KB
per example
The rack
What you have built
Export
This rack, as PyTorch
Every block knows its shapes and its weights, so it knows its own constructor. The result is a plain nn.Module with no dependency on Azimuth — paste it into your notebook.
The checker
What holds so far
60%
Shapes chain cleanly
Every block accepts what the one before it produces.
Produces the required output
Needs [B, 2] · rack currently ends at [B, 3, 128, 128]
Uses the blocks this idea needs
Still missing: Conv2D, Classifier head
Within the size budget
0 of 500K budget
Within the compute budget
0 of 50M budget
Shelf
Blocks available
Sources
Core layers
Reshaping
Normalization & regularization
Activations
Heads
Dials
No block selected
Pick a block on the rack to tune its dials and read what it does.