Different loss functions and up/downsampling in unet
smollm2-135M trained on a very small dataset
a small gpt2 setup
BPE tokenizer for Hindi
Resnet50 model trained on CIFAR100 with 74% val accuracy
