Scale AI

Fine-tuning LLMs

Fine-tuned custom-built LLMs by writing prompts and evaluating responses. Used the Hugging Face Transformers library to fine-tune models on specific tasks and datasets — including GPT-3.5, BERT, and T5 — to improve performance on targeted downstream tasks. Worked with the research team to identify weak spots, iterated on training data, and documented all experiments for future reference.