a set of utils based on LLMs
with psum one can get quick summaries from arXiv papers.
pass one paper identifier on the command line to get the summary of such a paper or pass multiple such identifiers to get a summary of multiple papers.
$> python llm_utils/psum.py --papers 2401.14887 2401.14887v3
Combined Summary:
This research analyzes the impact of Information Retrieval (IR) components on Retrieval-Augmented Generation (RAG) systems, which enhance traditional Large Language Models (LLMs) by incorporating external data. The study focuses on the optimal characteristics of a retriever for effective prompt formulation, specifically the type of documents to retrieve. The findings reveal that including irrelevant documents can increase performance by over 30% in accuracy, contradicting initial assumptions. This research highlights the need for specialized strategies to integrate retrieval with language generation models, paving the way for future research in this field.
it works both on OpenAI and HF APIs, HF is the default.
by default it uses a Mixtral-7B instructed model.
before running the script, make sure to export your HF token, e.g. export HUGGINGFACEHUB_API_TOKEN=123token456.