mukaj/Llama-3.1-Hawkish-8B

Warm
Public
8B
FP8
32768
License: other
Hugging Face
Overview

Llama-3.1-Hawkish-8B: Financial Domain Specialist

mukaj/Llama-3.1-Hawkish-8B is an 8 billion parameter model built on the Llama 3.1 architecture, specifically fine-tuned to excel in financial and economic reasoning. It leverages a newly generated dataset of 50 million high-quality tokens covering diverse financial topics such as Economics, Fixed Income, Equities, Corporate Financing, Derivatives, and Portfolio Management. To preserve general capabilities, this financial data was mixed with instruction sets on Coding, General Knowledge, NLP, and Conversational Dialogue.

Key Capabilities & Differentiators

  • Exceptional Financial Knowledge: The model has demonstrated a significant improvement in financial understanding, notably achieving a passing score on the CFA Level 1 exam, a benchmark known for its rigor.
  • Strong Performance in Economics & Math: Benchmarks show improved performance in these critical areas compared to the base model.
  • Outperforms Larger Models: In CFA Level 1 mock exams, Llama-3.1-Hawkish-8B has shown to outperform much larger models, including Meta-Llama Instruct 70B and Palmyra Fin 70B, in overall weighted average scores.
  • 32K Context Length: Supports processing longer financial documents and complex scenarios.

Ideal Use Cases

  • Financial Analysis: Answering complex questions related to economics, investment, and corporate finance.
  • Educational Tools: Assisting with learning and understanding CFA curriculum topics.
  • Research & Development: Exploring financial data influences in LLM training and performance.

Important Considerations

This is a research model derived from Meta's LLaMA 3.1 architecture and is subject to its community license terms. It is intended for academic and research purposes only and should not be used in production environments, for real-world financial decision-making, or advisory services. Users acknowledge its experimental nature and use it at their own risk.