Hugging Face Overview
Overview
Finance-Llama-8B is an 8 billion parameter language model, fine-tuned by tarun7r from the unsloth/Meta-Llama-3.1-8B base model. It leverages the Josephgflowers/Finance-Instruct-500k dataset, which comprises over 500,000 entries, to specialize in financial tasks. The model is designed for robust financial reasoning, question answering, and multi-turn conversational capabilities.
Key Capabilities
- Extensive Financial Coverage: Trained on a diverse dataset including financial QA, reasoning, sentiment analysis, topic classification, and multilingual NER.
- Multi-Turn Conversations: Capable of engaging in rich, contextually aware dialogues.
- Diverse Data Sources: Integrates data from Cinder, Sujet-Finance-Instruct-177k, Phinance Dataset, BAAI/IndustryInstruction_Finance-Economics, and Josephgflowers/Financial-NER-NLP.
- CFA Level 1 Performance: Demonstrates competitive performance on CFA Level 1 mock exams, achieving a 'PASS' result with a weighted average of 73%, outperforming Meta-Llama Instruct 8B and closely matching GPT-4o-mini and Meta-Llama Instruct 70B in specific sections.
Good For
- Financial Reasoning and Analysis: Ideal for tasks requiring deep understanding of financial concepts.
- Financial Chatbots and Assistants: Suitable for building conversational AI agents specialized in finance.
- Information Extraction: Effective for tasks like entity recognition and sentiment analysis within financial texts.
- Academic and Research Purposes: An experimental implementation for exploring the influence of financial data on language models.