This is your work, valued

Manchester, UK

Ali Naeimi

Advanced
@alint77

ML performance engineer currently focusing on LLM and proteinLM pretraining efficiency and distributed training infra. Open to positions in industry

nanogpt-fp8. Nanochat inspired LLM pretraining using Transformer-Engine with MXFP8 and NVFP4 support. Up to 30% faster than nanochat

6

matmul_assembly_x86. Hyper-optimized FP32 GEMM kernels in handwritten AVX2 ASM with a worklog of optimizations implemented

5

llm.c. 3x faster LLM training on CPU than Karpathy's original repo

2

flash-mHC. Fast implementation of manifold-constrained hyperconnections in Triton.

2

cmri-yolosam. Python

1

particle-simulator. C

1