In recent months, there has been a surge in the development of Large Language Models, including both proprietary closed-source models like GPT-4 and open-source models. This article aims to provide an example of how to build a tool for summarizing information, highlighting key technologies and ideas along the way. The goal is to create a tool that can pull from various data sources, run on personal devices, experiment with different configurations, and eventually scale up to a cluster. The article introduces the concept of fine-tuning, specifically the technique of Low Rank Adaptation (LoRA), which offers efficient parameter reduction for fine-tuning models. Several language models, such as FLAN-T5, Llama 2, and GPT-J 6B, are suggested as options for fine-tuning. The article also discusses various tool suites, including Hugging Face, Transformers, Accelerate, PEFT, Sematic, and Gradio, that can be used to manage, interface with, and fine-tune models for the summarization task. Finally, the article concludes with examples and results achieved using the developed pipeline.
https://www.sematic.dev/blog/tuning-and-testing-llama-2-flan-t5-and-gpt-j-with-lora-sematic-and-gradio