https://www.davidlang.tech/tags/fine-tuning/ Recent content in Fine-Tuning on David Lang Hugo en Wed, 15 May 2024 00:00:00 +0000 https://www.davidlang.tech/posts/fine-tuning-llms-when-and-how-to-customize-ai-models/ Wed, 15 May 2024 00:00:00 +0000 https://www.davidlang.tech/posts/fine-tuning-llms-when-and-how-to-customize-ai-models/ <p>Fine-tuning adapts a base model to your domain with labeled examples. Use it when prompting and RAG cannot achieve consistent style, format, or task-specific behavior.</p> <h2 id="when-to-fine-tune">When to Fine-Tune</h2> <ul> <li>Fixed output schema (legal clauses, medical codes)</li> <li>Brand voice across thousands of responses</li> <li>Specialized terminology poorly covered by general models</li> </ul> <h2 id="when-not-to-fine-tune">When Not to Fine-Tune</h2> <ul> <li>Facts that change frequently (use RAG)</li> <li>One-off tasks (use prompting)</li> <li>Small datasets without validation (risk overfitting)</li> </ul> <h2 id="openai-fine-tuning-flow">OpenAI Fine-Tuning Flow</h2> <p>Prepare JSONL with <code>messages</code> arrays. Upload, create job, evaluate on a holdout set. Monitor loss and human ratings before promoting to production.</p>