—Image input
[B, 3, 64, 64]no weights · 48 KB
Skip connections
A residual block that survives depth
Your plain convolutional stack got worse when you made it deeper — not overfit, just worse on the training set too. The 2015 answer was to change the shape of the block itself.
Build a convolutional block whose output is added back to an earlier tensor, then classify into 10 classes. The residual only connects when the shapes across the span match exactly.
The meter
What this architecture costs
[B, 64, 64, 64]Parameters
1.8K
FLOPs / example
7.1M
Activations
1.0 MB
per example
The rack
What you have built
[B, 3, 64, 64]Conv2D
[B, 64, 64, 64]1.8K params · 7.1M FLOPs · 1.0 MB
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
40%
Shapes chain cleanly
Every block accepts what the one before it produces.
Produces the required output
Needs [B, 10] · rack currently ends at [B, 64, 64, 64]
Uses the blocks this idea needs
Still missing: Residual add, Classifier head
Blocks are in a workable order
Conv2D must come before Residual add.
Within the size budget
1.8K of 3.0M 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.