6d ago · 11 min read · TLDR: Supervised fine-tuning (SFT) is the stage where a pretrained model learns task-specific response behavior from curated input-output examples. It is usually the first alignment step after pretraining and often the foundation for later RLHF. Good...
Join discussion6d ago · 12 min read · TLDR: Full fine-tuning updates every model weight, which is expensive in memory, compute, and storage. PEFT methods update only a small trainable slice. LoRA learns low-rank adapters on top of frozen base weights. QLoRA pushes efficiency further by q...
Join discussion6d ago · 12 min read · TLDR: Fine-tuning a 7B-parameter LLM updates billions of weights and requires expensive GPUs. LoRA (Low-Rank Adaptation) freezes the original weights and trains only tiny adapter matrices that are added on top. 90%+ memory reduction; zero inference l...
Join discussion
Mar 7 · 26 min read · Every tutorial on fine-tuning LLMs starts the same way: "Spin up an A100 on AWS..." Not this one. In this guide, you'll fine-tune Meta's Llama 3.2 1B into a Text-to-SQL assistant — entirely on your Ma
Join discussionMar 5 · 8 min read · Six months ago my team was building an internal support tool for a B2B SaaS company — about 120 employees, docs spread across Notion, Confluence, and a half-dead SharePoint instance from 2019. The ask was simple: a chatbot that could answer questions...
Join discussionMar 5 · 8 min read · Six months ago my team shipped a customer support bot that confidently told users our return window was 60 days. It's 30. The model had been fine-tuned on product documentation from 2023, and nobody — including me — thought to check whether that spec...
Join discussionMar 4 · 10 min read · # Fine-tuning vs RAG: When to Use Each Approach for Production LLMs You've shipped a proof-of-concept with GPT-4, your demo went well, and now engineering leadership wants it in production by next quarter. Then someone asks the question that keeps M...
Join discussionFeb 28 · 7 min read · TL;DR RAG is still the default for fast-changing knowledge, citations, and compliance-heavy use cases. Fine-tuning is for behavior, not your constantly changing knowledge base. Long context did not kill RAG; recent benchmarks show there is no univer...
Join discussion
Feb 27 · 7 min read · TL;DR RAG is still the default for fast-changing knowledge, citations, and compliance-heavy use cases. Fine-tuning is for behavior, not your constantly changing knowledge base. Long context did not kill RAG; recent benchmarks show there is no univer...
Join discussionFeb 21 · 6 min read · A pergunta certa antes de escolher a técnica Antes de falar de ferramentas, responda mentalmente: Você quer mudar comportamento ou conhecimento? O conhecimento precisa estar atualizado e auditável? Você precisa de consistência rígida de formato/esti...
Join discussion