Shiva Jaiswal

A detailed guide and implementation that explains using LoRA for parameter‑efficient fine‑tuning of GPT‑2 Medium (r=16, alpha=32), multi‑GPU training tips, hyperparameters, and results (validation perplexity, training time).

An essay examining why BM25 remains central to search pipelines, discussing practical BM25 implementation, speed/interpretability benefits, and hybrid BM25+neural re‑ranking architectures.