Overview

Finance-Llama-8B is an 8 billion parameter language model developed by tarun7r, fine-tuned from unsloth/Meta-Llama-3.1-8B. Its primary focus is on financial tasks, reasoning, and multi-turn conversations, leveraging a comprehensive dataset for specialized performance.

Key Capabilities

• Financial Specialization: Tailored for financial reasoning, question answering, entity recognition, and sentiment analysis.
• Extensive Coverage: Trained on over 500,000 entries from the Josephgflowers/Finance-Instruct-500k dataset, encompassing financial QA, reasoning, sentiment analysis, topic classification, and multilingual NER.
• Multi-Turn Conversations: Designed to handle rich dialogues with an emphasis on contextual understanding.
• Diverse Data Sources: Integrates data from various high-quality financial datasets, including Cinder, Sujet-Finance-Instruct-177k, Phinance Dataset, and BAAI/IndustryInstruction_Finance-Economics.

Performance

The model demonstrates competitive performance on CFA Level 1 mock exams. In a comparison against GPT-4o-mini, Meta-Llama Instruct 8B, and Meta-Llama Instruct 70B, Finance-Llama-8B achieved a weighted average of 73%, successfully passing the mock exam. This places its performance close to GPT-4o-mini (75%) and Meta-Llama Instruct 70B (70%), significantly outperforming Meta-Llama Instruct 8B (53%).

Intended Uses

This model is an experimental research implementation for academic and research purposes, exploring the influence of financial data in training language models. It is not intended for financial decision-making and users assume full responsibility for its application.