You have a deployment of an Azure OpenAI Service base model.

UTF-8 is the universal encoding standard: it is backward-compatible with ASCII for standard Latin characters, supports every Unicode code point covering over 140,000 characters, and is the de facto standard for all modern APIs including Azure OpenAI. This matters for fine-tuning because training data often contains diverse characters from multiple languages, special punctuation, and domain-specific symbols. If they use ASCII encoding, any accented character or non-Latin script will be corrupted or lost entirely. UTF-16 adds a Byte Order Mark and can cause parsing issues with tools that expect standard JSONL. ISO-8859-1 covers Western European characters only and fails immediately with CJK or Arabic scripts. The Azure OpenAI fine-tuning documentation explicitly states that training files must be UTF-8 encoded, making it the only safe choice.

Microsoft Learn Reference Topic: Azure OpenAI fine-tuning – File format and encoding requirements