What problem does it solve? Fine-tuning large language models on Google Cloud involves many error-prone steps: environment and IAM setup, dataset formatting and validation, model and region selection, cost estimation, job submission, monitoring, and deployment. This Skill guides an AI agent through the entire supervised tuning lifecycle on Agent Platform with guardrails that prevent common failures. ## Core Features & Use Cases - End-to-End Tuning Workflow: Covers environment setup, dataset preparation and validation, model and hyperparameter recommendation, job submission, monitoring, and deployment for both open models (Gemma, Llama, Qwen) and Gemini models. - Deterministic Helper Scripts: Includes scripts to convert and validate JSONL datasets, list available base models, estimate tuning cost, submit and cancel tuning jobs, and poll job status. - Use Case: A user wants to fine-tune Llama 3.1 8B on their own instruction data. The Skill validates the dataset, recommends hyperparameters from the models catalog, estimates cost, verifies the GCS dataset URI, and submits the job only after explicit user confirmation. ## Quick Start Ask the agent to fine-tune an open model on Agent Platform using your JSONL dataset and let it walk you through setup, configuration, and job submission.
