10 videos • 53 views • by Laxmi Kant | KGP Talkie Learn how to fine-tune LLMs on a Mac with Apple Silicon: no NVIDIA GPU, no cloud GPU, everything runs locally on a MacBook. This free, hands-on series fine-tunes small open-source language models (under 2B parameters) step by step with MLX, mlx-lm, Hugging Face Transformers, PEFT and TRL on PyTorch MPS, and Unsloth Studio.
Every lesson builds one working project and tests the model before and after fine-tuning:
• Text to SQL with Qwen 3.5 0.8B (LoRA with MLX)
• Messy invoices and receipts to clean JSON with IBM Granite
• Support-ticket classification with Google Gemma 3 270M in Unsloth Studio
• A customer-support chatbot in your brand's voice (SFT + DPO)
• Skill extraction from job posts (QLoRA)
• A prompt-injection firewall for your AI app
• A search-relevance judge for an online shop
• Bank-statement categorisation for a budget (DoRA)
• Teach a tiny model to reason with reinforcement learning (GRPO)
• Run your fine-tuned model on an iPhone with MLX Swift
You will learn dataset preparation, LoRA, QLoRA and DoRA settings, supervised fine-tuning (SFT), DPO, GRPO, evaluation, and how to run small LLMs locally.
Full course: https://kgptalkie.com/fine-tuning-llm