Show HN: Python Package for fine-tuning LLMs without writing code (github.com)

A new Python package called Upasak has been launched, designed for fine-tuning large language models (LLMs) without requiring users to write code. Built around Hugging Face Transformers, Upasak offers a no-code/low-code framework that allows users to fine-tune LLMs with an intuitive Streamlit-based interface. It supports various dataset formats and includes features for privacy compliance, such as built-in PII detection and data anonymization. Upasak caters to a wide audience, whether for educational purposes, rapid prototyping, or internal fine-tuning on sensitive data. Significantly, Upasak lowers the barrier to entry for developers and researchers who may lack deep expertise in machine learning but wish to leverage LLMs in their applications. Its support for both full parameter and LoRA fine-tuning methods ensures flexibility, making it easy to experiment with different training configurations. Future updates are expected to expand its capabilities to additional model families, including multimodal variants. The easy integration into existing pipelines and automated processes, such as generating customized inference scripts post-training, enhances its utility for practical development and research.