- ๐ M.Sc. student in Intelligent Systems (AI / NLP) at Nagoya University, Sasano Lab โ MEXT Scholar
- ๐ฌ Research: mechanistic interpretability of LLMs y
- ๐ Research Intern at Sakana AI, Tokyo
- โก Also writing GPU kernels โ CUDA, Triton, HIP on NVIDIA and AMD hardware
- ๐ซ Reach me at hamdi.abderrahmene.t6@s.mail.nagoya-u.ac.jp
| Project | What it is |
|---|---|
| 100 Days of GPU Kernels โญ 600+ | 100 consecutive days writing and benchmarking CUDA, Triton and HIP kernels in public โ GEMM, softmax, fused MMA-ReLU, Flash Attention, on both NVIDIA and AMD. |
| Native Sparse Attention in C & CUDA | From-scratch implementation of Native Sparse Attention in pure C and CUDA, validated against a reference PyTorch version, with an accompanying write-up. |
Languages
ML & Interpretability
GPU & Parallel Computing
Systems & MLOps
Data
Arabic (native) ยท English (full professional) ยท Japanese (JLPT N1) ยท French (professional)