March 15th 2026 at 12:00
Extending TrivialAugument with a transform grid and a tinygrad-native deformation.
Read this
February 22nd 2026 at 12:00
Extending the training set by doing random transforms.
Read this
February 8th 2026 at 12:00
The JIT problem has been resolved.
Read this
January 27th 2026 at 12:00
Suspecting tinygrad JIT is at fault.
January 25th 2026 at 12:00
Decreasing sample count to see if the network overfits as expected.
January 18th 2026 at 12:00
Trying out Focal loss with multiple parameter sets.