Ankitgoelvirgo·6d agoBuilding a Zero-Cost JSON Extractor: A Fun Local LLM Fine-Tuning POC on Mac M4 ProIf you’ve ever tried using a foundational LLM to extract structured data into JSON, you know the struggle. No matter how much you beg in…
Thendral Magudapathi·Oct 12LoRA Under the Hood: Low-Rank Factorization of Weight UpdatesUnderstanding low-rank adaptation through matrix shapes, initialization, and a numerical example.
TheCraftedDev·Oct 3Why LoRA and QLoRA Make Fine-Tuning CheaperImagine you have a large AI model with billions of numbers inside it. These numbers are called weights, and they help decide how the model…
Grace Esther S.·Sep 17LoRA vs QLoRA vs Prefix Tuning: Which PEFT Method Actually Wins?A practical comparison of three fine-tuning strategies, and what the numbers really mean for deployment
Bhavana·Sep 16We Fine-Tuned the Model. It Got Better at Our Task and Quietly Worse at Everything Else.Not a dramatic outage, not a data leak. Here’s a representative version of how this plays out in production:
jaytank·Sep 2LoRA and QLoRA Explained: The Low-Rank Math Behind Efficient Fine-TuningWhy full-parameter model updates waste millions in GPU compute - and how training lightweight adapter matrices enables fine-tuning 70B…
Grnitish·Aug 17My Weekend Learning on Fine-Tuning and Quantizing LLMsI’ve relied on Medium a lot over the years, especially when prepping for interviews and getting up to speed on new concepts in the LLM…
Akash Chandrasekar·Aug 13I Thought Fine-Tuning Was the Starting Point. Turns Out, I Was Missing the Whole Story.Understanding the LLM training pipeline, why fine-tuning exists, and where techniques like SFT, LoRA, QLoRA, and RLHF actually fit.A response icon1A response icon1
Kaush·Aug 13I Fine-Tuned a Model With QLoRA and It Lost to a Rule-Based BaselineI fine-tuned a model with QLoRA, 4-bit, LoRA adapters, the works. And when I compared the fine-tuned model to the rule-based baseline I had…A response icon1A response icon1
