Ml-Systems

A visualization of INT4 quantization layers inside a transformer model, showing weight matrices being compressed and dequantized.

Building a From-Scratch Group-Wise INT4 Quantization Layer for LLaMA Weights

A hands-on guide to implementing group-wise INT4 quantization for LLaMA-style transformers, covering per-group scale calibration, fused dequantization kernels, and benchmarks that demonstrate real ML-systems engineering skill.

A visual representation of low-rank adaptation matrices decomposing a large weight matrix into two smaller ones.

Build a LoRA Rank Adaptation System From Scratch

A hands-on build guide for a LoRA rank adaptation system from scratch in PyTorch, with production-grade architecture, runnable code, and a roadmap to senior-level extensions.

Cover image showing INT4 quantization flow: FP16 weights compressed to INT4 with block-wise scales

Build an INT4 Block-Wise Weight-Only Quantizer for Transformer Attention Layers

Build a production-grade INT4 block-wise weight-only quantizer from scratch in pure Python. This guide covers calibration, dequantization, benchmarking, and why this project signals senior ML-systems skill to hiring managers.

Diagram of a small GPT training loop with mixed precision and gradient accumulation.

Building nanoGPT From Scratch: A Portfolio-Worthy Deep Learning Project

Ship a portfolio-grade nanoGPT implementation in pure PyTorch: a tight training loop with AMP, gradient accumulation, cosine LR scheduling, and weight tying — the exact patterns production teams use.

Layered diagram of a quantization pipeline showing calibration, packing, and fused dequantize-matmul.

Build a GPTQ int4 Quantization Engine in Python/NumPy

Ship a runnable int4 weight-only quantization engine from scratch: GPTQ-style calibration, group-wise affine scales, and a fused dequantize-matmul kernel. Built to demonstrate real ML systems skill on a portfolio.