The-FinAI/PIXIU

This repository introduces PIXIU, an open-source resource featuring the first financial large language models (LLMs), instruction tuning data, and evaluation benchmarks to holistically assess financial LLMs. Our goal is to continually push forward the open-source development of financial artificial intelligence (AI).

What it solves

PIXIU addresses the lack of comprehensive, open-source resources for developing and evaluating Large Language Models (LLMs) specifically for the financial domain. It provides a unified framework that combines a specialized model, a diverse instruction dataset, and a rigorous evaluation benchmark to move beyond simple NLP tasks and include real-world financial predictions.

How it works

PIXIU is composed of three primary components:

  • FinMA: A financial LLM fine-tuned on specialized instruction data to provide learning and prediction capabilities for financial tasks.
  • FIT: A multi-task, multi-modal financial instruction dataset used to train and fine-tune LLMs.
  • FinBen: A comprehensive evaluation benchmark that tests models on financial language understanding and prediction across various contexts.

Who it’s for

This project is designed for AI researchers and developers working on financial NLP, quantitative analysts, and anyone building specialized LLMs for the finance industry.

Highlights

  • Multi-modal Data: Includes not only text (reports, news, tweets) but also time series data for stock movement prediction.
  • Broad Task Coverage: Covers sentiment analysis, classification, knowledge extraction, number understanding, text summarization, credit scoring, and forecasting.
  • Open Resources: Openly releases models, instruction tuning data, and evaluation datasets on Hugging Face.
  • Multilingual Support: Includes enhanced versions of the benchmark supporting English, Chinese, and Spanish.

Related

  • Project
    microsoft/typeagent-py

    typeagent is an experimental Python package (requires Python 3.13+) aiming to bring structured RAG functionality from a TypeScript project to Python. It's still a prototype, sends input to a third-party LLM (so don't use it with confidential data), and installs via `pip install typeagent`.

  • Project
    Deodat-Lawson/LaunchStack

    A TypeScript engine for building AI-native applications that provides a modular, ports-based architecture for RAG, document ingestion, and LLM orchestration.

  • Project
    ThariqS/ai-newtab

    AI Homepage is a Chrome extension that uses your Anthropic API key to send a brief summary of your recent browsing (up to 8 pages) to Claude, which then generates a personalized new‑tab page in any style you describe.

  • Project
    bpwhelan/GameSentenceMiner

    GSM is a tool that automates the creation of language-learning flashcards by capturing game audio, screenshots, and text directly while you play.