biba10/Generative-Cross-lingual-ABSA ? reverse-engineered prompt

Build me a Python project for cross lingual aspect based sentiment analysis that can train and evaluate on multiple languages, especially English, Czech, Spanish, French, Dutch, Russian, and Turkish. I want one main script for standard training with a sequence to sequence model like T5, plus separate entry points for open source LLM prompting, ChatGPT prompting, and instruction tuning. It should read the language data from a data folder with train, dev, and test files, let me choose source and target language, task type, batch size, epochs, learning rate, and whether to use few shot examples or constrained decoding.

Please make it run from the command line, save the best checkpoint when I use a validation set, and report the main ABSA scores clearly. If you need anything uncertain, check the current docs online and make sensible defaults.