Coursework implementations for the Advanced Deep Learning module at HHU.
The repository tracks each exercise as a notebook, with commits split into provided material and my implementation work.
| Exercise | Topic |
|---|---|
| 1 | Speculative decoding |
| 2 | Speculative decoding evaluation and performance measurements |
| 3 | GPT KV cache for autoregressive generation |
| 4 | Triton fused add-ReLU kernel and benchmarking |
| 5 | Triton LayerNorm forward/backward |
| 6 | Wavefront scheduling for stacked RNNs |
| 7 | Mamba2 block for character-level language modeling |
Large generated artifacts, datasets, checkpoints, and local runtime files are excluded from version control. The notebooks are intended as implementation artifacts rather than polished lecture notes.