jhon53/Llama3_1_8B_Finance_QLoRA-merged-16bit
jhon53/Llama3_1_8B_Finance_QLoRA-merged-16bit is an 8 billion parameter Llama 3.1 Instruct model, fine-tuned by jhon53 (Dharmik Bhingradiya) using QLoRA for 3-class financial sentiment classification. This model excels at identifying negative, neutral, and positive sentiment in financial texts, achieving significantly improved accuracy and Macro-F1 scores on financial benchmarks like FPB and FiQA-SA compared to its base model. It is optimized for financial sentiment analysis and research in finance-focused LLMs.
Loading preview...
Overview
This model, jhon53/Llama3_1_8B_Finance_QLoRA-merged-16bit, is a specialized fine-tuned version of the meta-llama/Meta-Llama-3.1-8B-Instruct base model. Developed by Dharmik Bhingradiya, it focuses on 3-class financial sentiment classification (negative, neutral, positive).
Key Capabilities & Features
- Financial Sentiment Classification: Specifically trained to analyze and classify sentiment in financial texts.
- QLoRA Fine-tuning: Utilizes QLoRA with a 4-bit NF4 frozen base model and BF16 LoRA adapters for efficient fine-tuning.
- Enhanced Performance: Demonstrates substantial improvements over the base Llama 3.1 8B Instruct model on financial benchmarks:
- FPB Accuracy: Improved from 89.08% to 97.48%.
- FPB Macro-F1: Increased from 87.65% to 97.25%.
- FiQA-SA Accuracy: Rose from 81.20% to 94.02%.
- FiQA-SA Macro-F1: Grew from 67.05% to 83.35%.
- Training Data: Fine-tuned on the
FinGPT/fingpt-sentiment-traindataset, comprising approximately 76,000 examples.
Intended Use Cases
This model is designed for:
- Performing financial sentiment analysis.
- Research and experimentation with finance-focused Large Language Models.
- Classifying financial news, statements, and market-related text.
- Building finance-oriented NLP and Generative AI applications.