mirror of
https://github.com/MLSysBook/TinyTorch.git
synced 2026-05-24 17:27:11 -05:00
d181b1615f
Created test_cnn_integration.py with: ✅ Conv2d Operations Tests: - Verifies actual convolution (not just shape manipulation) - Edge detector test proves Conv2d computes correctly - Shape transformations for various configurations - Parameter count verification (448 params for 3→16, k=3) ✅ Pooling Operations Tests: - MaxPool2d actually computes maximum values - AvgPool2d actually computes averages - Shape transformations validated - Handles negative values correctly ✅ Numerical Stability Tests: - Zero inputs handled correctly - Negative values in pooling work properly ⚠️ Gradient Flow Tests (TODO): - Placeholder for Conv2d backward support - Will add when Conv2d autograd integration is implemented All forward pass tests passing (8/8)! These tests ensure CNNs actually work, not just shape shuffle.