LLM - Abrarqasim Blogs

LLM Quantization Flipped: A 4-Bit Model Beat Its 16-Bit Source

LLM Quantization Flipped: A 4-Bit Model Beat Its 16-Bit Source

A 4-bit model just beat the 16-bit checkpoint it was quantized from. I break down quantization-aware healing, the numbers, and what it changes for local…

Rayyan | August 31, 2026 | 8 min

Fine-Tuning a ColBERT Model on One GPU Just Got Practical

Fine-Tuning a ColBERT Model on One GPU Just Got Practical

Sentence Transformers v6 makes ColBERT-style retrieval trainable on one consumer GPU. The numbers, two traps in the fine print, and when it beats dense embeddings.

Rayyan | August 31, 2026 | 6 min

Few-Shot Prompting: When More Examples Make It Worse

Few-Shot Prompting: When More Examples Make It Worse

Adding more few-shot examples can quietly make your LLM prompt worse. Here is what the over-prompting research found and how I size the example set…

Rayyan | July 30, 2026 | 7 min

SLM vs LLM in 2026: When a Small Model Is the Right Call

SLM vs LLM in 2026: When a Small Model Is the Right Call

SLM vs LLM in 2026: why most of your model calls do not need a frontier model, the small-first cascade pattern that cuts cost, and…

Rayyan | July 28, 2026 | 7 min

RAG Chunking Strategies in 2026: What I Actually Use

RAG Chunking Strategies in 2026: What I Actually Use

After two years of running RAG in production, here are the chunking strategies I actually use, the ones I dropped, and why chunk size is…

Rayyan | May 4, 2026 | 8 min